Briefing CS-2026-069

Bluesky's API Servers Under Fire: The 313 Team DDoS Attack of April 2026


Late on April 15, 2026, attackers began flooding Bluesky's infrastructure with malicious traffic. What followed was roughly 24 hours of cascading API server failures, a pro-Iran hacktivist group taking credit on Telegram, a follow-up attack on Mastodon five days later, and a growing realization that even platforms built on open, decentralized protocols are only as resilient as the centralized servers sitting underneath them.

Bluesky is a decentralized social media platform built on the AT Protocol — an open-source framework designed so that users retain ownership of their data and identity, and anyone can in theory run their own infrastructure node. The project was initiated in 2019 inside Twitter by then-CEO Jack Dorsey, who funded a small research team and chose the name; it became an independent public benefit corporation in 2021 with software engineer Jay Graber as the first CEO. Graber led the platform through its growth phase before stepping down to become Chief Innovation Officer on March 9, 2026, with Toni Schneider — founding CEO of Automattic (the company behind WordPress.com) and a partner at True Ventures — taking over as interim CEO while the board runs a search for a permanent replacement. According to coverage from Recorded Future News, Bluesky had attracted approximately 43.7 million users by the time of the April 2026 incident, with Limelight Digital's statistics tracker placing it at roughly 43.5 million as of April 6, 2026 — the small delta reflects continued daily registrations. The incident hit at precisely the moment Bluesky's growth narrative was strongest, and it exposed a gap between the promises of decentralized architecture and the operational realities of running a platform at scale.

A Note on Bluesky's Leadership and Origin Story

Some outlets describe Bluesky as Jack Dorsey's platform or credit Dorsey as the founder; others credit Jay Graber, because Graber led the actual independent company that built and launched the product. Both framings are partially correct: Dorsey initiated the concept and provided initial Twitter funding, but he left the Bluesky board in 2024 and publicly distanced himself from the platform's moderation direction. Graber was CEO from 2021 to March 2026 and remains Bluesky's largest shareholder. At the time of the April 2026 DDoS, Toni Schneider — Automattic's founding CEO and a True Ventures partner whose firm was already an investor and advisor to Bluesky — was serving as interim CEO, and COO Rose Wang (who was Graber's San Francisco housemate while Graber was auditioning for the original CEO role) was the executive publicly attributing the incident to a cyberattack. We identify Wang as COO rather than listing a specific named CEO because the interim arrangement was in effect and public reporting consistently attributes the incident confirmation to Wang.

The attack also landed in a particularly charged geopolitical moment. One day before the DDoS began, on April 14, Russian internet regulator Roskomnadzor reportedly added Bluesky to its registry of banned websites, pushing Russian users who had already been cut off from Telegram, WhatsApp, Signal, Discord, and other foreign platforms toward whatever remained accessible. Twenty-four hours later the platform was under sustained attack from an entirely different direction.

What Happened: The Attack Timeline

Bluesky's team received the first reports of intermittent app outages at approximately 11:40 p.m. PDT on April 15, 2026, per the company's own official statement. Engineers worked through the night and into the following day. Chief Operating Officer Rose Wang publicly attributed the instability to a coordinated cyberattack, and Bluesky issued an official statement acknowledging that the incident was impacting operations across feeds, notifications, threads, and search.

A Note on Conflicting Start Times in the Reporting

TechCrunch, Fast Net Mon, and several other outlets report the attack "originally started on April 15 at around 8:40 p.m. ET" — while Bluesky's own statement places the first intermittent outage reports at 11:40 p.m. PDT on April 15. These are not the same moment. 11:40 p.m. PDT April 15 converts to 2:40 a.m. ET April 16, roughly six hours after TechCrunch's 8:40 p.m. ET figure. We follow Bluesky's official timestamp for when the team was formally alerted, while noting that external monitoring and downstream reporting observed user impact earlier that evening. Engadget separately reported that status-page issues began at 1:42 a.m. ET April 16, which is consistent with an attack that had ramped up over the preceding hours and reached a detection threshold at Bluesky around 11:40 p.m. PDT. The short version: both timeframes describe the same incident at different stages of detection.

The disruption was not a clean outage where the platform simply went dark. Instead, users encountered a fragmented and unpredictable experience that in some ways was more disorienting than a full blackout. Personal feeds sometimes loaded. Popular algorithmic feeds like Discover returned rate-limit errors. Profile pages would fail to load and require repeated refresh attempts. The platform's own status page was itself hit hard enough at points that users could not use it to track progress on the incident — a failure mode that deserves more attention than it typically receives. When a status page shares infrastructure dependencies with the production systems under attack, or when supporting services are indirectly degraded during a broad attack response, the monitoring layer that users rely on for ground truth about the incident becomes unreliable at precisely the moment demand for it peaks. In Bluesky's case, the status page intermittent failures meant users during the worst periods could neither access the platform nor confirm whether the disruption was platform-wide or local to their connection. The checklist item in the defender section of this briefing addresses this directly; the underlying principle is that status infrastructure must be isolated from production architecture at the networking and hosting layer, not just logically separated. In a major attack, any shared dependency is a shared failure point. In a sign of the operational pressure the team was under, one status update contained a visible typo referring to "one of our reginos" — a misspelling of "regions" that went uncorrected for hours.

Bluesky protocol engineer Bryan Newbold remarked around 3:46 a.m. ET that the services were getting hit hard, a candid observation from inside the incident that matched what users were experiencing from outside. Technical monitoring recorded service failures at approximately 1:42 a.m. ET Thursday per Engadget's reporting, with degradation escalating through the early-morning hours and continuing into the day.

April 4-6, 2026
Bluesky experiences a non-security-related outage on April 6 that takes roughly half of users offline for approximately eight hours, caused by ephemeral port exhaustion on internal memcached connections. Systems engineer Jim Calabro publishes a detailed engineering post-mortem on April 10. This is infrastructure strain context, not a DDoS.
April 14, 2026
Russia's Roskomnadzor reportedly adds Bluesky to its registry of banned websites, per reporting from Recorded Future News citing Russian digital rights organization RKS Global.
April 15 — 11:40 p.m. PDT
Bluesky's team receives the first reports of intermittent app outages. Engineers begin working through the night to diagnose and mitigate the incoming flood.
April 16 — early morning
313 Team posts on Telegram claiming a massive cyberattack targeting the Bluesky API, saying the attack would continue for additional hours and predicting ongoing disruption.
April 16 — 1:42 a.m. ET
Technical monitoring records service failures on the official Bluesky status page, per Engadget's reporting. Feeds, notifications, threads, and search degrade intermittently across the platform.
April 16 — 7:47 p.m. ET
Bluesky publicly confirms the cause as a sophisticated DDoS attack that intensified throughout the day. Statement from COO Rose Wang cites no evidence of unauthorized access to private data.
April 16 — 9 p.m. PDT
Bluesky reports the application has stabilized, though attack traffic continues in the background. Blacksky — an independent AT Protocol community — reports a spike in migration requests from affected Bluesky users. StatusGator's external monitoring later logs the April 16 outage window at roughly 17 hours 17 minutes total downtime.
April 17 — 4:36 p.m. ET
Bluesky confirms continued stability despite ongoing DDoS attempts, reiterates no evidence of data access, and announces the next update would arrive by the morning of April 20. StatusGator logs an additional 1 hour 31 minutes of downtime on this date.
April 20 — ~7:00 a.m. ET
Mastodon's flagship server (mastodon.social) is hit by a DDoS attack in a similar pattern. Mastodon posts its first status update at ~7 a.m. ET and deploys countermeasures; access is restored by 9:05 a.m. ET. No threat actor publicly claimed responsibility for the Mastodon attack. Security Affairs, TechCrunch, and FastNetMon all explicitly note the absence of any verified claim. Hackread attributed the Bluesky attack to 313 Team but made no verified attribution claim for the Mastodon incident.
April 20 — 1:08 p.m. ET
Bluesky publishes what it initially frames as its final status update, reporting the app had remained stable since April 16 evening with no evidence of unauthorized access.
April 20 — afternoon ET
Despite the "final update" post, Bluesky is hit by an additional DDoS attack on April 20 afternoon. Bluesky's status account reports "elevated errors and timeouts on some Bluesky-hosted services" and its status page briefly goes offline. The company's April 20 update-three blog post explicitly acknowledges the additional attack but notes the application remained largely stable throughout.

Throughout the incident, Bluesky consistently reported no evidence of unauthorized access to private user data — directly addressing the recurring concern that a sustained DDoS can serve as a distraction or smokescreen for a concurrent intrusion attempt.

What Did Mitigation Actually Look Like?

Bluesky's public communications described its team working through the night to mitigate the attack, but the company disclosed no specific technical mitigations at any stage of the incident. No information was released about upstream traffic filtering, CDN engagement, anycast routing changes, rate-limit threshold adjustments, or any other defensive measures applied during or after the event. What the public record does show is behavioral evidence: FastNetMon's coverage of the incident noted that the "Rate Limit Exceeded" errors visible to users were consistent with throttling controls being applied during the mitigation window, and that intermittent availability suggested active traffic management prioritizing some endpoints over others. The gradual stabilization pattern — personal feeds recovering before high-traffic feeds, the status page remaining intermittently available — is consistent with selective endpoint protection as mitigation resources were applied in stages. But none of that was confirmed by Bluesky. This absence of technical disclosure creates a gap in the public record that matters: security practitioners looking to learn from the incident have no confirmed information about what protective controls were in place before the attack, which of those controls failed first, or what was added during the incident response window.

Compounding this gap is the absence of a post-incident technical write-up from Bluesky for the DDoS event. When Bluesky's April 4-6 outage caused by internal port exhaustion resolved, systems engineer Jim Calabro published a detailed public post-mortem within four days, documenting the root cause, the specific service behavior that triggered the failure, the remediation applied, and what monitoring would be added to prevent recurrence. No comparable technical post-mortem has been published for the April 15-20 DDoS incident. The contrast is notable: the internal engineering failure received full public transparency; the externally driven security incident did not. This may reflect the different legal, reputational, and attribution-complexity considerations involved in DDoS post-mortems compared with infrastructure post-mortems, but the result is a publicly documented security incident with no confirmed forensic or technical learning record. NIST SP 800-61r3 (Incident Response Recommendations and Considerations for Cybersecurity Risk Management, 2025) — which supersedes SP 800-61r2 — addresses the integration of incident response into continuous cybersecurity risk management and emphasizes that lessons learned should be identified and shared as soon as they are available, not deferred until after recovery concludes.

A Note on "24-Hour Attack" Framing vs. Multi-Day Activity

Headlines across SecurityWeek, Security Affairs, Hackread, and SQ Magazine frame this as a "24-hour attack," because the period of acute user-visible disruption was roughly one day — from late April 15 PDT through the evening of April 16 PDT when the platform stabilized. That framing is accurate for the worst of it. But Bluesky's own status posts acknowledged ongoing DDoS attempts through at least April 17, and the company confirmed an additional DDoS attack on the afternoon of April 20 in its own blog post. The total attack window was therefore closer to five days of intermittent pressure, with one concentrated ~24-hour peak. We use both frames in this briefing because both are true of different slices of the incident.

In its official status post on April 16, Bluesky stated that its team had worked through the night to mitigate what it described as a sophisticated DDoS attack that intensified throughout the day. — Bluesky (@bsky.app), April 16, 2026

DDoS as Smokescreen

Security teams should always treat a sustained DDoS as a potential cover operation. Volumetric traffic floods logs, saturates monitoring pipelines, and diverts analyst attention — creating ideal conditions for a concurrent intrusion, data exfiltration, or credential stuffing campaign operating under the noise floor. Bluesky's confirmation that no user data was accessed is significant, but incident responders should never take that assessment as final until a complete forensic review of the attack window is complete.

Reference AT Protocol: High-Value API Endpoints for DDoS Targeting

The AT Protocol Lexicon is public. These are the XRPC endpoints with the highest server-side compute cost per request — the endpoints an attacker planning a slow-path Application Exhaustion Flood would target first. Each is formally documented at atproto.com/lexicons. Click any row to expand the defender note.

Endpoint
Cost tier
Why it matters
Aggregates the algorithmic Discover feed by querying the full social graph. Each request requires multi-step ranking computation across the BGS firehose. This is the highest-compute endpoint in the App View stack and the most damaging to flood — one malicious request here consumes far more resources than dozens of simple lookups. The degradation of Discover before personal feeds during the April 2026 attack is consistent with disproportionate App View load on this endpoint.
Assembles the personal timeline by querying all follows in the user repository, fetching their posts from the BGS firehose, and applying recency and relevance ordering. Compute cost scales with follow count — an attacker using high-follower seed accounts in a botnet maximizes per-request resource consumption significantly above the per-account average.
Fetches the full thread tree for a post, including parent and reply chains. Thread depth is unbounded by the protocol; deeply threaded viral posts require recursive graph traversal. Flooding this endpoint with requests targeting high-reply threads (easily identified from public feed data) produces disproportionate backend load per request volume.
Searches across the full indexed post corpus. Full-text search is among the most compute-intensive operations in any social graph. The AT Protocol search index is shared infrastructure — flooding it with complex queries degrades search availability platform-wide, not just for the requesting client. Broad or wildcard-style queries are especially expensive per call.
Creates a new authenticated session (login). Auth endpoints are typically rate-limited less aggressively than data endpoints to avoid locking out legitimate users. Flooding createSession with credential attempts consumes authentication server resources and — if rate limits are hit — locks real users out of their accounts. This is also the primary target for credential stuffing campaigns running in parallel with a volumetric flood as a cover operation.
Resolves a handle (e.g. user.bsky.social) to a DID (Decentralized Identifier). Individually cheap, but called by clients on every profile render. Under flood conditions, resolution failures cascade into broken profile pages and notification failures platform-wide. High request volume from many distributed sources produces a long-tail degradation pattern that is harder to detect than a single-endpoint flood precisely because no individual source exceeds typical thresholds.

Attribution: The 313 Team Claim

Bluesky declined to name an attacker. In response to press questions, the company said it was not in a position to speculate about attribution — a standard posture for an active incident where forensic work is still in progress and where false flags are routine. But a claim of responsibility appeared on Telegram within hours of the attack's escalation, and multiple threat intelligence outlets including SecurityWeek, The Record, Hackread, and Security Affairs subsequently attributed the incident to the group making the claim.

The claimed actor is a pro-Iran hacktivist collective calling itself the 313 Team, also known as the Islamic Cyber Resistance in Iraq. According to 313 Team's own Telegram post, as reported by Threat Beat citing the group's channel directly, the group announced on April 16 that they had carried out a cyberattack targeting the API of the Bluesky social networking platform, and claimed the attack "peaked at 1 terabyte." The group also initially predicted the attack would last three hours, then publicly revised that estimate upward to approximately 10 additional hours as Bluesky continued posting status updates.

A Note on the "1 Terabyte" Claim

Several outlets including Threat Beat quoted 313 Team's Telegram post verbatim as claiming the attack "peaked at 1 terabyte." Strictly speaking, terabytes are a unit of data volume, not attack rate — the meaningful DDoS metric is bits or packets per second. If the group meant 1 terabit per second (1 Tbps), that would still be substantial but nowhere near the current Cloudflare-documented ceiling of 31.4 Tbps from the Aisuru-Kimwolf botnet (Q4 2025). If they meant 1 terabyte of cumulative traffic, that's a far less dramatic claim than it sounds. Neither Bluesky nor any independent telemetry provider has published the actual volumetric measurement, so the "1 terabyte" figure should be read as unverified attacker self-reporting rather than confirmed peak bandwidth. We include it because it was the specific claim made, and readers will see it repeated; we flag the unit ambiguity because anyone evaluating this as capability data needs to understand the caveat.

Who Is the 313 Team?

The 313 Team is a pro-Iran hacktivist group that, per its own GitHub organization page, describes itself as founded in June 2023 and self-identifies as the cyber training and software development arm of what it calls the Islamic Cyber Resistance Axis. The group operates primarily out of Iraq and targets organizations or platforms associated with countries it views as supporting the United States or Israel. Cybersecurity researchers have linked 313 Team to retaliatory cyber operations aligned with the interests of Iran-backed militias, and the group has been assessed by multiple outlets including SecurityWeek as active during the ongoing conflict dynamics between the United States, Israel, and Iran.

Two details about the 313 Team profile rarely appear in incident-focused coverage. First, the name's origin: the number 313 is a direct reference to the Muslim warriors who fought at the Battle of Badr in 624 CE — a decisive early Islamic military victory — a symbolic choice the HawkEye threat advisory notes the group shares with other resistance-aligned forces in the region. The character "313" recurs across Iran-aligned and Resistance Axis branding as a shorthand for a smaller force defeating a larger enemy against the odds, which is precisely the self-image the group projects. Second, the group's proof-of-attack methodology is rigidly consistent: after each claimed operation, 313 Team posts CheckHost.net or similar uptime verification screenshots to its Telegram channel alongside the target URL, a timestamp, and a claimed duration. This is not incidental — it is a standardized evidence package designed for Telegram amplification and downstream intelligence collection by other groups within the coalition. Security teams monitoring hacktivist channels should treat the arrival of a CheckHost.net screenshot from this group as the formal attribution signal, not the initial claim text. The screenshots are the group's preferred credibility mechanism.

In the months leading up to the Bluesky incident, the group had claimed a lengthy list of DDoS operations: Microsoft 365 disruption in mid-March 2026, the X platform at the end of March, Cinemark theater servers, Amazon Prime Video, Dropbox (reportedly deflected by Dropbox's security), Yahoo, AOL, the Clever education platform, Bahraini government portals, Kuwaiti government infrastructure (reportedly 26 government IP domains over 72+ hours), and Saudi Arabia's Absher citizen services platform. The pattern is consistent: short-duration, high-visibility DDoS operations against well-known platforms, typically announced on Telegram with CheckHost.net uptime screenshots as proof, followed by claims of impact that often exceed what independent reporting confirms.

An important but rarely cited dimension of the 313 Team profile comes from an exclusive interview the group gave to Daily Dark Web in April 2026. In that exchange, the group provided insight into its ideological framing and future intentions that goes well beyond its public Telegram activity. The group stated that it operates across all fields with members specialized in DDoS attacks, experienced hackers, and skilled programmers — and described its operations as a deliberate extension of physical conflict into cyberspace. On the question of targeting civilians, the group invoked an explicit proportionality doctrine, citing the Iranian school in Minab as a justification for targeting the Clever and Moodle educational platforms. This context matters for security professionals evaluating the group's target selection logic: attacks are not random. They are framed by the group as calibrated responses to specific events, which means target selection shifts as the geopolitical situation evolves.

The same interview revealed the group's stated capability trajectory. 313 Team claimed to be in the final development stages of RATs, ransomware, and stealer malware. The group also stated intentions to target supply chains using advanced software — a significant escalation from the disruptive-but-limited DDoS operations that define its current profile. Whether those claims reflect genuine capability or aspirational positioning, they establish a threat posture that security teams should factor into any long-term assessment. A hacktivist group that successfully expands from Layer 7 DDoS into persistent access tools represents a qualitatively different risk category.

In the Daily Dark Web interview, the group also stated its intention to expand its technical capabilities, promising to increase the destructive capacity of future DDoS operations and bring new offensive software into its arsenal. — 313 Team, exclusive interview with Daily Dark Web, April 2026

A further layer that rarely surfaces in incident-focused coverage: Unit 42 and GNET Research have documented that 313 Team's operations since late February 2026 operate within a structured coordination architecture. The "Electronic Operations Room," established February 28, 2026 following Operation Epic Fury, functions as a primary coordination hub for Iran-aligned hacktivist operations. Research by SecurityScorecard's STRIKE team, which analyzed 250,000 messages from 178 active groups during the preceding 12-day Israel-Iran conflict, found that attack timings and target selection patterns across these collectives were consistent with institutional IRGC orchestration rather than organic activism. 313 Team sits within this ecosystem. The group publicly denies direct IRGC or Iraqi Popular Mobilization Forces ties, but independent analysts note the ideological alignment is tight and the operational tempo tracks with IRGC interests. This context matters for attributing motive: the Bluesky attack on April 15-16 came at a moment when Iran had been largely offline for over 47 days due to its own internet blackout, and its proxy hacktivist infrastructure was operating at elevated tempo on its behalf.

One further escalation signal that defenders in critical sectors should not overlook: while 313 Team's current profile is DDoS-and-defacement, the broader Iran-aligned hacktivist ecosystem in which it operates has already crossed into industrial control systems. Unit 42's April 2026 threat brief documented that a cluster tracked as CL-STA-1128 — also known as Cyber Av3ngers or Storm-0784 — shifted in late March 2026 from its historic focus on internet-connected Unitronics PLCs to targeting Rockwell Automation FactoryTalk software, installing it on VPS infrastructure to enable exploitation of operational technology systems. That cluster operates within the same broad Iran-aligned operational umbrella. This is not evidence that 313 Team is targeting OT systems; it is evidence that the ecosystem surrounding 313 Team already is. Security professionals evaluating 313 Team risk in isolation are evaluating a narrow slice of the actual threat picture. The group's current DDoS capability sits within a coordinated network that has demonstrated, at the ecosystem level, the capability and intention to move from web-layer disruption into industrial control system exploitation.

Why Bluesky? The Missing Stated Motive

One question the 313 Team coverage largely skips: why Bluesky specifically? Per Heise Online's reporting, the group did not explain its motivation for the DDoS attacks against Bluesky or Mastodon, but merely reported alleged successes. This absence of stated grievance is itself informative. 313 Team's targeting logic does not appear to require a specific political trigger tied to the target's own conduct; it appears to follow a visibility threshold. Bluesky had grown to over 43 million users largely as a result of migrations away from X following Elon Musk's political positioning and the re-election of Donald Trump, making it a high-profile Western communication platform by early 2026. Per The Record's reporting, 313 Team targets organizations associated with countries seen as supporting the United States or Israel — Bluesky's profile, its association with the mass migration from Musk's X, and its rapid growth narrative as a Western-aligned social alternative likely made it a sufficiently visible target without requiring any platform-specific grievance. Security professionals evaluating 313 Team's target selection should not expect explicit justifications; the group tends to announce outcomes rather than publish rationales, and the absence of stated motivation does not mean the targeting was random.

On April 20, 2026 — five days after the Bluesky attack began — Mastodon reported that its flagship server mastodon.social was hit by a DDoS attack. Mastodon deployed countermeasures and restored access within approximately two hours. The timing and target selection, one decentralized microblogging network immediately followed by another, tracks with the 313 Team's hopping pattern across social platforms. Hackread attributed the Mastodon incident to 313 Team explicitly; Security Affairs reported that no threat actor had publicly claimed responsibility for the Mastodon attack at time of writing. This is a genuine reporting disagreement in public sources, so we note both views rather than assert a single attribution.

The decentralized nature of the Fediverse is a true advantage.

— Andy Piper, Mastodon head of communications, via TechCrunch, April 2026
Attribution Caveat

Hacktivist claims of responsibility should be treated as one input among several, not as ground truth. SecurityWeek, Security Affairs, and other outlets have noted that 313 Team's claims are not independently verified, that the group has at times exaggerated attack duration and impact, and that hacktivist personas are sometimes used by state-linked operations as attribution fronts. 313 Team itself said the Bluesky attack was expected to last three hours, then extended that estimate; the actual concentrated disruption ran closer to 24 hours, and intermittent follow-on attempts continued through April 20. Bluesky's own refusal to speculate on attribution reflects appropriate incident response discipline, not a dismissal of the claim.

Why API Servers Are the Target

313 Team's own Telegram statement described the operation as targeting Bluesky's API — not its website, not its mobile client, but the application programming interface layer that sits between everything else. Understanding why API servers bore the brunt of this attack requires understanding what those servers actually do. Every interaction on a modern social platform — loading a feed, fetching a user profile, posting content, sending a notification — is mediated by API calls. The API layer is not a back-office system; it is the functional core of the entire user experience. When it goes down, or begins throttling under load, nothing works. This is precisely what makes APIs such a high-value target for denial-of-service campaigns.

For Bluesky specifically, the AT Protocol architecture means the API surface is particularly expansive. The protocol is designed for interoperability, which requires a rich and publicly accessible set of endpoints. Third-party developers, custom feed builders, and alternative client applications all make calls to Bluesky's infrastructure. That openness, which is one of the platform's primary selling points, simultaneously increases the number of potential request vectors an attacker can leverage. Public API documentation is a gift to developers and an attack roadmap to adversaries at the same time.

What this means technically: AT Protocol's API layer uses a purpose-built remote procedure call system called XRPC (Cross-organizational Remote Procedure Calls). Every single endpoint in the network is addressed via a path prefixed with /xrpc/ followed by a Namespaced Identifier (NSID) — for example, /xrpc/app.bsky.feed.getTimeline or /xrpc/com.atproto.server.createSession. The schemas for every one of these endpoints are defined by Lexicon, AT Protocol's global schema language, and the full set of Lexicon definitions is publicly published and versioned. In August 2025, portions of the AT Protocol specification were submitted to the Internet Engineering Task Force (IETF) as a Birds of a Feather proposal, and formal Internet Drafts were submitted covering repository formats and the protocol's general architecture. The implication for defenders: this is not merely a well-documented API — it is becoming IETF-standardized documentation, meaning the complete technical specification of every requestable endpoint, its parameter types, its authentication requirements, and its expected load characteristics is available as structured reference material to any attacker who knows where to look. An adversary planning a targeted slow-path attack against Bluesky's compute-heavy endpoints did not need to reverse-engineer the application; the attack surface is published.

The AT Protocol's three core service layers introduce a further structural consideration. Every user-facing request flows through three distinct Bluesky-operated components: the Personal Data Server (PDS), which hosts user repositories and routes requests; the Big Graph Server (BGS), which crawls all PDS instances and outputs a single unified event firehose that all downstream services consume; and the App View, which ingests that firehose and aggregates content into the social graph users see. All three of these components are operated by Bluesky the company in the production network, regardless of the protocol's theoretical support for third-party operators. An attacker who understands the architecture knows that overwhelming the App View layer — the component that aggregates data from the firehose into algorithmic feeds like Discover — produces the most user-visible damage per request, because every algorithmic feed request requires significantly more computation than a simple object lookup. The visible failure mode during the attack, in which high-traffic algorithmic feeds degraded before personal feeds, is consistent with the App View bearing a disproportionate share of the attack load.

Architecture AT Protocol Request Flow: Where the Attack Hit
CLIENT User / 3rd-party app XRPC requests PDS Personal Data Server identity + repo BGS Big Graph Server event firehose APP VIEW feed aggregation PRIMARY TARGET API GATEWAY /xrpc/* endpoints rate limiting firehose BOTNET FLOOD API FLOOD HTTP 429 rate-limit errors surface to all legitimate clients
Click a layer label above to explore each component and how it factored into the attack. Hover the App View node to see why it was the highest-value target.

The Rate Limit Error as a Weapon

The "Rate Limit Exceeded" messages users saw throughout the outage were not bugs. They were the expected behavior of a system under attack, surfacing as user-facing errors. Rate limiting — the practice of capping how many requests a given client can make in a defined time window — is a fundamental API defense mechanism. When an attacker floods endpoints with requests, the server returns HTTP 429 errors to clients that have crossed the threshold, regardless of whether those clients are legitimate users or botnet nodes. During the Bluesky attack, the incoming flood was sufficient to push legitimate users into the same rate-limit buckets typically reserved for abuse cases, effectively locking them out of high-traffic feeds and endpoints.

Popular algorithmic feeds like Discover are particularly vulnerable to this pattern because they aggregate content from across the network and require more server-side computation per request. A single malicious request to a compute-heavy endpoint consumes disproportionately more resources than a simple profile lookup. Attackers who understand a platform's API topology can deliberately target the most resource-intensive endpoints to maximize damage per request — a technique sometimes called a slow-path attack at the application layer.

Layer 7 Attacks Are Harder to Stop

Traditional network-layer DDoS mitigation filters traffic by volume and source IP. Layer 7 attacks — those that target the application layer and make structurally valid HTTP requests to API endpoints — are significantly harder to differentiate from legitimate traffic. A Layer 7 flood can use valid credentials, correctly formatted requests, and distributed source IPs drawn from botnet infrastructure, making it appear indistinguishable from organic traffic spikes at the network layer. Akamai's 2026 Apps, APIs, and DDoS State of the Internet report, published March 17, 2026, found Layer 7 DDoS attacks rose 104% over the preceding two years, driven by DDoS-for-hire services and rentable botnets.

The Scale of the Threat in 2026

The Bluesky incident did not occur in isolation. According to Cloudflare's 2025 Q4 DDoS Threat Report published in February 2026, DDoS attacks surged 121% in 2025, with the company automatically mitigating an average of 5,376 attacks every hour across its network. Network-layer DDoS attacks more than tripled year-over-year, and Cloudflare mitigated a record-setting 31.4 terabits per second attack from the Aisuru-Kimwolf botnet in Q4 2025 — part of a campaign the company named "The Night Before Christmas" that targeted Cloudflare customers and Cloudflare's own dashboard with hyper-volumetric HTTP floods exceeding 200 million requests per second.

Akamai's 2026 report found that 87% of organizations experienced an API-related security incident in 2025, and the average daily API attacks per organization rose 113% year-over-year. Web application attacks climbed 73% between 2023 and 2025. The DDoS-as-a-service industry has matured to the point where a capable, sustained attack against a mid-sized platform requires neither technical sophistication nor significant financial outlay. Botnet operators lease access to attack infrastructure by the hour. The obfuscation layers inherent in these services — IP rotation, encrypted command-and-control channels, residential proxy pools — make attribution extremely difficult, which is precisely why hacktivist claims of responsibility require corroboration rather than acceptance.

Context Attack Scale in Context: Putting the 1 TB Claim on the Map

313 Team claimed the Bluesky attack peaked at "1 terabyte." No independent volumetric telemetry was published. This comparison places that unverified claim against confirmed industry benchmarks to calibrate the claim's plausibility.

313 Team claimBluesky attack, April 2026 (unverified)
~1 Tbps claimed
Cloudflare recordAisuru-Kimwolf botnet, Q4 2025
31.4 Tbps
GitHub DDoS (2018)Memcached amplification attack
1.35 Tbps
Cloudflare avg hourly5,376 attacks/hr mitigated in 2025
varies
Typical DDoSaaS burstcommodity rental botnet, 2026
100-300 Gbps
Note: 313 Team's claim uses "1 terabyte" as a data-volume unit, not a bandwidth-rate unit (Tbps). The mapping above treats it as roughly equivalent to a peak bandwidth figure for comparison purposes only. The claim is unverified and cannot be confirmed from public telemetry. The Cloudflare 31.4 Tbps figure is confirmed published data. Bar widths are proportional to a log scale for visual readability.
Reference Attack Indicators, Actors, and Framework Mapping

Key indicators, threat actor profile, techniques, and framework references associated with the Bluesky DDoS incident and the broader API attack landscape.

Pro-Iran hacktivist collective, active since June 2023; claimed responsibility for the Bluesky attack via Telegram on April 16, 2026; also claimed the April 20 mastodon.social attack. The group describes itself as ideologically aligned with Iran's Wilayat al-Faqih framework, and has publicly denied direct organizational ties to the IRGC or Iraqi Popular Mobilization Forces — though independent analysis by SecurityScorecard and Unit 42 places 313 Team within a coordinated Electronic Operations Room established February 28, 2026.
313 Team-affiliated development arm hosting the HackBar browser-based security audit tool on GitHub; indicates web application attack tooling beyond volumetric DDoS. The group has publicly stated it is in the final stages of developing RATs, ransomware, and stealer malware — claims that, if operationalized, would represent a significant escalation from DDoS and defacement operations.
User-facing error surfaced across Bluesky feeds, profiles, and Discover during the attack window — indicator of volumetric API endpoint flooding.
Key AT Protocol API servers experienced extended unavailability starting approximately 1:42 a.m. ET April 16 (Engadget), contributing to cascading feed and notification failures. StatusGator external monitoring logged 17 hours 17 minutes of downtime on April 16, plus an additional 1 hour 31 minutes on April 17.
Bluesky's status.bsky.app experienced its own disruptions during the attack, indicating the attack volume was sufficient to impair monitoring infrastructure. This is operationally significant: when monitoring goes down alongside production, incident response fidelity degrades and parallel intrusion activity is harder to detect.
313 Team attribution methodology: Telegram posts accompanied by third-party uptime verification screenshots of target unreachability. This is a consistent pattern across the group's operations; the screenshots constitute their standard proof-of-disruption artefact, rather than technical forensic evidence of attack origin.
Adversary use of volumetric traffic to overwhelm network or service availability; maps to the sustained flood targeting Bluesky's API infrastructure. Bluesky's public characterization of the attack as "sophisticated" is consistent with multi-vector methodology combining network-layer and application-layer vectors.
Application-layer exhaustion targeting specific service endpoints; consistent with the observed degradation pattern on Discover and high-traffic feeds.
Sub-technique: flooding application-layer resources with requests that consume disproportionate compute, consistent with targeting of resource-intensive API endpoints such as Bluesky's Discover feed aggregator.
Pre-incident resource development; maps to hacktivist groups leasing DDoS-for-hire botnets and coalition-pooled infrastructure for volumetric operations. In an exclusive interview with Daily Dark Web, 313 Team stated they intend to increase the power of their DDoS attacks and expand their technical arsenal — context relevant to defenders assessing future capability trajectory.
Sub-technique of T1498: one or more systems generate high-volume traffic directly toward the target to saturate network capacity. Relevant to the botnet-powered flood characterization of the Bluesky attack; maps to the 313 Team claim of a "1 terabyte" peak and the coalition-pooled DDoS infrastructure described in SecurityScorecard and Unit 42 reporting.
Sub-technique of T1499: adversary sends high volumes of requests to exhaust web server or network service resources. Directly maps to the HTTP flood pattern against Bluesky's API layer that produced widespread HTTP 429 rate-limit errors, degraded Discover feed aggregation, and generated intermittent failures on the status page.
Pre-attack reconnaissance sub-technique: gathering target DNS records and passive DNS data to map infrastructure before operations begin. For Bluesky and AT Protocol, this reconnaissance requirement is largely eliminated — the complete XRPC endpoint specification, authentication requirements, and rate-limit behavior are formally published in the AT Protocol Lexicon and submitted as IETF Internet Drafts, meaning an attacker can enumerate compute-heavy endpoints from documentation alone without active scanning.
Finalized June 2025. Directly applicable to the Bluesky incident: addresses API risk factors across the development and runtime lifecycle, rate limiting at the API gateway layer, API inventory and discovery controls for shadow and deprecated endpoints, and protection measures for publicly exposed API endpoints. The document provides both basic and advanced controls organized by API lifecycle phase, making it the primary NIST reference for organizations assessing API-layer DDoS resilience after this class of incident. Published by NIST SP 800-228 (Chandramouli and Butcher, 2025). See: csrc.nist.gov/pubs/sp/800/228/final
Resilient Interdomain Traffic Exchange: BGP Security and DDoS Mitigation. Covers the network-layer mitigations most relevant to the T1498/T1498.001 vectors in this incident: remotely triggered black hole (RTBH) filtering, Flowspec rules for upstream traffic filtering, unicast Reverse Path Forwarding (uRPF) to prevent IP spoofing, and source address validation. A public draft of Revision 1 was released January 2025 (public comment period closed February 25, 2025). See: csrc.nist.gov/pubs/sp/800/189/final
Current incident response guidance (supersedes SP 800-61r2). Applicable to the Bluesky response and post-incident transparency gaps identified in this briefing: structured incident handling, parallel forensic review during active DDoS events, and integration of incident response into broader cybersecurity risk management aligned to NIST CSF 2.0. The absence of a technical post-mortem for the DDoS event — in contrast to the detailed April 10 engineering write-up for the April 4-6 port exhaustion incident — reflects a gap this framework is designed to address. See: csrc.nist.gov/pubs/sp/800/61/r3/final
Security and Privacy Controls for Information Systems and Organizations. SC-5 (Denial-of-Service Protection) requires organizations to protect against and limit the effects of DoS attacks, including by employing rate limiting, traffic filtering, and capacity management. SC-5(3) specifically addresses managing excess capacity and bandwidth to reduce the susceptibility to DoS. SI-10 (Information Input Validation) supports rejection of malformed API requests at the gateway layer. These controls map directly to the multi-layer API DDoS resilience posture described in the defender section of this briefing.
Cloudflare 2025 Q4 data documents multi-vector attacks combining Layer 3/4 and Layer 7 techniques; Bluesky's characterization as "sophisticated" is consistent with multi-vector methodology.
Akamai 2026 SOTI data shows Layer 7 DDoS up 104% over two years; DDoSaaS services enable low-cost, high-volume attacks with built-in attribution obfuscation via IP rotation and residential proxy pools.
Group leverages Telegram channels for attack announcements, coordination with coalition partners, and psychological amplification of reputational damage. 313 Team also uses AI-generated propaganda imagery across Telegram — burning city imagery, Islamic iconography, and multilingual threat messages — coordinated with SEPAHCYBERY (an IRGC-linked channel) and broader pro-Iranian media infrastructure, per GNET Research analysis.

The Decentralization Paradox

One of the more instructive dimensions of this incident is what it reveals about the gap between decentralized architecture in theory and in practice. Bluesky is built on the AT Protocol, which allows any operator to run their own Personal Data Server and participate in the broader network. In theory, this means there is no single point of failure. In practice, Bluesky the company operates the dominant infrastructure — the main PDS, the primary API servers, the feed generators, and the appview — and the overwhelming majority of users rely entirely on that infrastructure.

This was visible during the outage. Third-party communities built on the same AT Protocol, including Blacksky — a network that runs its own independent infrastructure — continued to function throughout the attack. According to reporting from TechCrunch, Blacksky's team confirmed a significant spike in migration requests from Bluesky users during the incident window as developers and users sought alternatives. But for ordinary users without the technical background to migrate their identity to a different PDS, those alternatives were largely theoretical. The protocol may be decentralized; the user base is not.

The Mastodon incident on April 20 provides an instructive contrast on exactly this point. Mastodon's flagship mastodon.social server was disrupted, but the many smaller instances that make up the broader Mastodon/Fediverse network were unaffected. Mastodon's head of communications explicitly cited decentralization as the reason many users never noticed. The structural difference: the Fediverse is federated at the user level, with tens of thousands of independent instances. AT Protocol is federated at the architectural level, but in practice the user base has concentrated on Bluesky's servers.

Comparison Decentralization in Theory vs. Under Attack

AT Protocol is architecturally decentralized. Bluesky's user base is operationally centralized.

Theory
Any operator can run a Personal Data Server. Users hold portable identities. No single point of failure mandated by the protocol design.
Reality (April 2026)
43+ million users concentrated on Bluesky's servers. App View, BGS, and primary PDS all operated by one company. Migration possible in theory; effectively inaccessible during active outage for typical users.
What worked
Third-party operators (e.g., Blacksky) running independent AT Protocol infrastructure remained fully operational. Developers with technical fluency could and did migrate.
What failed
Ordinary users had no practical path to an alternate server during the outage. Identity portability and outage resilience are not the same property. The protocol's open API surface documented the attack targets.
Outcome
Decentralized at the protocol level, centralized at the user level. The attack surface matched the centralization, not the theory.

mastodon.social was disrupted April 20. The broader Fediverse was not. Federation worked as intended.

Theory
Thousands of independently operated instances. No canonical central server. Users federated at the account level across distinct operators.
Reality (April 20, 2026)
mastodon.social went down. Smaller instances across the Fediverse remained fully available. mastodon.social restored within approximately two hours. Most Fediverse users never noticed the attack.
Why it held
Real user distribution across thousands of instances. Disrupting one node degrades that node's users but does not cascade to the broader network. No shared central API layer serving the whole user base.
Attribution note
Hackread attributed the Mastodon attack to 313 Team. Security Affairs reported no public claim at time of writing. Both positions remain in the public record.
Outcome
Decentralized at the protocol level and at the user level. Operational resilience matched the architecture.

The architectural label is less important than where users actually live and what they can actually do during an incident.

For platform architects
Protocol-level decentralization does not guarantee operational resilience. Model your threat surface against actual user distribution, not theoretical node topology.
For security practitioners
"Decentralized" on a product page does not translate to distributed attack surface. Evaluate where the API gateway, feed aggregator, and authentication services actually run and who operates them under pressure.
For open-protocol platform operators
The same documentation transparency that builds developer trust provides adversaries with a pre-built attack surface map. A public XRPC Lexicon with compute-cost signals requires more aggressive rate-limit tuning on resource-intensive endpoints, not standard flat-rate limiting.

What Could Affected Users Actually Do?

During the outage, a question surfaced that points directly at the gap between AT Protocol's theoretical promises and the practical experience of its user base: could affected users migrate to another server and continue operating normally? Technically, yes — the protocol is designed for exactly that. In practice, for the overwhelming majority of Bluesky's 43-plus million users, migration was either unknown as an option or operationally inaccessible during an active outage. TechCrunch reported that Blacksky confirmed a significant spike in migration requests from developers, power users, and technically sophisticated participants in the AT Protocol ecosystem. But migration during an outage requires knowing that alternative Personal Data Servers exist, knowing how to request a migration, having the technical background to complete it, and doing all of that while the platform infrastructure you need to initiate the process is partially or fully unavailable. For a typical user, those requirements effectively mean migration was not a live option during the attack window. The AT Protocol's identity portability is a real architectural property. But identity portability is not the same as resilience. A protocol feature that is difficult to discover, technically complex to execute, and entirely unavailable to initiate during an outage does not function as a meaningful user protection against availability incidents. This gap between the protocol's capabilities and users' ability to act on them is a design and communication problem as much as a technical one — and the Bluesky incident makes it visible in a way that normal operations never would.

This is not unique to Bluesky. It is a structural challenge that every protocol-based open network faces. The convenience of a well-funded central operator attracts the user base, and the user base's concentration on that operator creates the centralization the protocol was designed to prevent. From an attacker's perspective, this is ideal: the protocol's open nature provides a large and publicly documented API surface, while the user base's concentration on a single operator means a successful attack against that operator achieves maximum disruption.

What This Means for Platform Security

For security professionals evaluating social platforms or API-dependent services, the Bluesky incident is a reminder that "decentralized" is an architectural description, not a security guarantee. The attack surface must be evaluated against what infrastructure users are actually concentrating on — not what the protocol theoretically allows. API threat modeling should account for the centralization patterns that emerge from user behavior even in nominally open systems.

API Security Risk Landscape in 2026

The Bluesky attack lands in the middle of the worst period for API security on record. Akamai's 2026 State of the Internet report, released March 17, 2026, found that 87% of organizations experienced an API-related security incident in 2025, and that average daily API attacks per organization surged 113% year-over-year. Web application attacks rose 73% globally between 2023 and 2025. The threat surface has expanded dramatically as AI-assisted development has accelerated API proliferation, often without commensurate security review — Akamai specifically flagged risks from "vibe coding," or software built quickly with AI assistance, as a source of vulnerabilities and misconfigurations reaching production without adequate testing.

The OWASP API Top 10 in Context

NIST published SP 800-228, Guidelines for API Protection for Cloud-Native Systems, in June 2025. It addresses API risk identification and controls across the full API lifecycle, including rate limiting at the gateway layer, API inventory management to surface shadow and deprecated endpoints, and protection measures for publicly exposed APIs. For organizations conducting post-incident API security reviews following an event like the Bluesky DDoS, SP 800-228 provides the current authoritative NIST framework for structuring those controls.

While the Bluesky attack was a volumetric DDoS rather than an exploitation of a specific vulnerability, the incident sits within a broader attack landscape that encompasses the full OWASP API Security Top 10. Broken Object Level Authorization (BOLA) remains the leading category of exploitable API flaw across industry reporting. Misconfigured authentication mechanisms, excessive data exposure, and broken function-level authorization account for the largest share of confirmed API breaches. These vulnerability classes are relevant to any post-incident analysis because they represent the exploitation paths attackers may attempt once volumetric traffic has degraded logging and monitoring fidelity.

Shadow APIs and zombie APIs — undocumented or deprecated endpoints that remain accessible — represent a particular risk category for social platforms that have evolved rapidly. Any endpoint not included in an active inventory cannot be covered by rate limiting, schema validation, or WAF rules. Bluesky's rapid growth trajectory increases the likelihood that endpoint sprawl has outpaced security governance in at least some areas of the stack, and the platform had publicly acknowledged earlier April 2026 outages attributed to "port exhaustion" from accommodating its user growth.

A Note on "Port Exhaustion" vs. DDoS Reporting

TechBooky and other outlets noted that Bluesky experienced earlier outages in April 2026 attributed to "port exhaustion" — a capacity issue in which a server runs out of available ephemeral TCP ports to service new connections. Bluesky systems engineer Jim Calabro published a detailed public post-mortem on April 10, 2026 documenting the relevant incident: a Saturday April 4 paging alert followed by a Monday April 6 outage that took roughly half of Bluesky's users offline for about eight hours, caused by a new internal service that launched 15,000 to 20,000 goroutines per request and exhausted ephemeral ports on localhost memcached connections that were accumulating in TIME_WAIT state. This is not the same thing as a DDoS. Port exhaustion in this case was a scaling/resource-tuning problem driven by internal workload patterns; a DDoS is a deliberate external flood. We mention the earlier port-exhaustion incident because it indicates Bluesky's infrastructure was already under stretch capacity nine days before the 313 Team attack landed, which is operationally relevant context. It is not evidence of a cyberattack, and no outlet has suggested the April 4-6 incident was DDoS-related.

Broken Authentication Under Load

Authentication failures are also more likely to occur during high-load events. When API servers are under DDoS pressure, authentication and session validation services may be throttled or degraded to preserve capacity for core functions. This creates a window during which authentication checks are less rigorous, token validation may be skipped or cached in ways that introduce risk, and anomalous access patterns may not trigger the same alerts they would under normal conditions. A sophisticated attacker who combines a DDoS campaign with a targeted credential stuffing or session hijacking operation during the degraded window is exploiting both the technical and the human dimensions of incident response simultaneously.

In the Bluesky case, there is no indication this occurred. But the pattern is well-documented in other incidents, and the risk should be factored into any organization's DDoS response playbook.

Defense Insight: Identity-Aware Rate Limiting

Basic rate limiting by IP address is insufficient against distributed attacks. Modern API security requires identity-aware throttling that ties limits to authenticated sessions, API keys, or behavioral fingerprints rather than source IPs alone. Behavior-based rate limiting — which flags anomalous request patterns even within nominal thresholds — is increasingly essential against low-and-slow attacks that stay below per-IP limits while still generating destructive aggregate load across distributed botnet nodes.

Defender Actions and Mitigations

The Bluesky incident offers concrete lessons for security teams responsible for protecting API-dependent platforms. The following actions address both the immediate DDoS threat and the broader API security posture exposed by this class of attack.

Edge-based DDoS mitigation should be treated as a baseline, not an enhancement. Deploying protection at the network edge — through a CDN or dedicated scrubbing service — means malicious traffic is blocked before it reaches origin infrastructure. An attack blocked in a distributed edge node in the region closest to its source consumes no origin server resources. Bluesky's own experience demonstrates that when edge mitigation fails to absorb or filter an incoming flood, the degradation cascades through every layer of the stack — from API servers to notification pipelines to the status page itself. NIST SP 800-189 (Resilient Interdomain Traffic Exchange: BGP Security and DDoS Mitigation) provides the technical framework for upstream network-layer DDoS controls: remotely triggered black hole (RTBH) filtering, Flowspec rules, unicast Reverse Path Forwarding (uRPF), and source address validation (SAV). A public draft of Revision 1 was released in January 2025.

Rate limiting must be implemented at multiple layers. IP-level limits at the network perimeter handle the most obvious volumetric patterns. API-key-level and session-level limits at the gateway layer address authenticated abuse. Endpoint-specific limits — with tighter thresholds on compute-heavy operations like feed aggregation — reduce the damage from targeted slow-path attacks against resource-intensive routes. Schema validation at the API gateway, which rejects any request that does not conform to the documented API contract, blocks malformed requests before they reach the application layer.

AT Protocol's own reference implementation provides a technically precise window into how this rate limiting is designed to work. The @atproto/xrpc-server package exposes a rate limits configuration supporting three categories: global IP-based limits applied across all methods, shared rate limit pools shared across multiple methods, and per-method limits defined inline. The reference examples in the published library show a global IP limit of 3,000 points per 5-minute window, and per-method limits of 100 requests per hour and 1,000 requests per day for write operations. The backing store for production rate limiting is Redis (via the RedisRateLimiter class), not in-memory state — meaning that under high-concurrency attack conditions, the Redis layer itself can become a latency bottleneck and a secondary point of failure, independent of the API servers. For security engineers implementing similar architectures, the Redis rate-limit tier is a component that must be provisioned and monitored at the same resiliency tier as the API servers it protects, not treated as lightweight support infrastructure.

API inventory and surface area management are non-negotiable. An organization cannot rate-limit or monitor an endpoint it does not know exists. Automated API discovery should be run continuously to surface shadow endpoints, deprecated versions still accepting traffic, and third-party integrations that have expanded the attack surface. For platforms using open protocols like AT Protocol, the documented API surface is a gift to attackers as much as it is to developers — every publicly available API specification is also an attack roadmap.

Defenders with visible geopolitical exposure should be monitoring hacktivist Telegram channels and threat intelligence feeds for early targeting signals. 313 Team's operational pattern includes announcing targets shortly before or during attacks, giving defenders a short but sometimes actionable window. Organizations with perceived ties to US, Israeli, or Gulf-aligned interests should factor pro-Iran hacktivist targeting into current risk assessments, and communication platform dependencies should be mapped and backup channels identified before an outage, not during one.

Defender Checklist API DDoS Resilience Controls

Verify these controls are in place for any externally facing API infrastructure. Click each item to track completion.

0 / 7 completed

What's Disputed in the Reporting, At a Glance

This incident generated substantial coverage very quickly, and a number of specific details vary across sources. Because CyberSpit readers will encounter other accounts that differ from this one in small but real ways, the table below captures the main points where public reporting is not uniform — and which version we went with and why. The goal is not to adjudicate who's right; it's to make our editorial choices legible so readers can evaluate our work against the broader reporting environment.

Clarifications Where Public Reporting Diverges, and Our Rationale

Points where coverage varies across outlets, with the version used in this briefing and the reasoning behind it.

What's out there
TechCrunch and others: April 15 around 8:40 p.m. ET. Bluesky's own post: 11:40 p.m. PDT April 15 (= 2:40 a.m. ET April 16).
Our position and why
Use Bluesky's official time for when the company was formally notified, and note the external observation was earlier. The two are not contradictory — they describe different detection thresholds as the attack ramped.
What's out there
Many headlines: "24-hour attack." 313 Team's claim: 3 hours, revised upward. Bluesky: ongoing DDoS through April 17, additional attack April 20.
Our position and why
Describe the concentrated ~24-hour disruption, but flag that intermittent activity persisted through April 20. Both framings are true of different parts of the incident.
What's out there
313 Team: "peaked at 1 terabyte" per their Telegram post (Threat Beat). No independent volumetric telemetry published.
Our position and why
Report the claim as the attacker made it, but flag that terabytes measure data volume, not attack rate, and that the claim is unverified. Do not present it as confirmed peak bandwidth.
What's out there
Hackread, Security Affairs, SecurityWeek, Heise, The Record all report the 313 Team Telegram claim. Bluesky publicly declined to attribute.
Our position and why
Present the claim as made and contextualized (consistent with 313 Team's operational pattern); present Bluesky's non-attribution as appropriate incident-response discipline. Neither confirms nor dismisses the claim.
What's out there
Hackread: 313 Team claimed the April 20 mastodon.social attack. Security Affairs: no actor had publicly claimed responsibility.
Our position and why
Acknowledge both public positions explicitly. The timing and target-selection pattern fits 313 Team, but the public-claim evidence is split, so we do not assert a single attribution for the Mastodon incident.
What's out there
Recorded Future and Hackread: 43.7 million (April 2026). Limelight Digital: 43.5 million (April 6, 2026). Sprout Social/Backlinko: 40.2 million (late 2025).
Our position and why
Cite 43.7 million as the figure used in the specific April 2026 incident reporting, note the 43.5 million tracker number, and explain the delta is continuous daily growth rather than a factual dispute.
What's out there
Some coverage credits Jack Dorsey as founder; others credit Jay Graber.
Our position and why
Describe both accurately: Dorsey initiated the concept inside Twitter and chose the name; Graber was the first CEO of the independent company and led the build. Do not pick one as the founder.
What's out there
Much reporting refers to Graber as Bluesky CEO. In fact, Graber stepped down March 9, 2026 to become Chief Innovation Officer, with Toni Schneider (Automattic founding CEO, True Ventures partner) serving as interim CEO while the board searches for a permanent replacement.
Our position and why
Do not name a specific named CEO for the incident window beyond noting Schneider's interim role. Use COO Rose Wang for incident-related executive attribution, since public reporting consistently attributes the cyberattack confirmation to Wang.
What's out there
Some outlets reference earlier April 2026 Bluesky outages linked to port exhaustion as related reliability context.
Our position and why
Mention the earlier April 4-6 outage (documented in Jim Calabro's public engineering post-mortem published April 10) as context for infrastructure strain, but explicitly note it was an ephemeral-port resource issue caused by an internal service pattern, not a DDoS. Conflating it with the 313 Team attack would misrepresent both incidents.
What's out there
Headlines: "24-hour attack." StatusGator external monitoring: 17h 17m on April 16 plus 1h 31m on April 17. Bluesky: "stable since evening of April 16" but additional attack on April 20.
Our position and why
Use "roughly 24 hours" for the concentrated acute period. Cite the StatusGator figures separately as the most precise external telemetry. The measured-downtime numbers and the "24-hour" headline figure are measuring slightly different things — one is strict unavailability, the other is total incident window.
What's out there
Engadget: started at 1:42 a.m. ET. Some secondary coverage: 2:42 a.m. ET.
Our position and why
Use Engadget's 1:42 a.m. ET figure as the earliest precise external timestamp. Secondary figures appear to be downstream rewrites with a transcription drift.
What's out there
Some coverage explicitly ties the attack to the Russia-Bluesky block one day prior. Others treat them as independent events.
Our position and why
Note the timing adjacency as context because it is objectively true. Do not claim direct causation between the Russian block and the Iran-linked 313 Team operation — no source has established such a link, and the two actors have very different motivations.
Knowledge Check Test Your Understanding of This Incident

Four questions drawn directly from the analysis above. Answers reveal the reasoning, not just the correct option.

Question 1 of 4
Why did Bluesky's Discover feed degrade faster than personal feeds during the attack?
Bluesky's App View aggregates content from the full event firehose to build algorithmic feeds. A single request to getTimeline or getFeed triggers significantly more server-side computation than a profile fetch or post lookup. Attackers targeting the most compute-expensive endpoints produce maximum disruption per request — the pattern Akamai calls an application-layer slow-path attack. The degradation order users observed during the incident is consistent with the App View bearing a disproportionate share of the flood load.
Question 2 of 4
What made Bluesky's API surface particularly easy for attackers to enumerate without any active reconnaissance?
AT Protocol uses a publicly versioned schema system called Lexicon that defines every XRPC endpoint, its parameter types, authentication requirements, and expected behavior. In August 2025 the specification was submitted to the IETF as Internet Drafts. This means an adversary planning a targeted attack can identify compute-heavy endpoints — such as app.bsky.feed.getTimeline — from official documentation alone, without scanning, probing, or any active pre-attack activity. The transparency that builds developer trust simultaneously removes the reconnaissance barrier for attackers.
Question 3 of 4
Why should every sustained DDoS against an API platform trigger a parallel investigation beyond the DDoS itself?
A high-volume DDoS creates exactly the conditions that benefit a concurrent intrusion attempt: logging throughput degrades, authentication log fidelity drops, on-call engineers are focused on the flood, and monitoring infrastructure may itself be impaired. Bluesky found no evidence of unauthorized data access during the attack window — but that determination requires active investigation. The absence of confirmed parallel intrusion is a result of investigation, not an assumption that can be made by default. This is a core principle in post-DDoS incident response: treat availability attacks as potential cover events until the auth logs and data pipelines are independently reviewed.
Question 4 of 4
313 Team provided no specific grievance against Bluesky. What does the available evidence suggest drove target selection?
Per reporting from The Record, 313 Team targets organizations associated with countries perceived as supporting the US or Israel. The group rarely publishes explicit rationales; it announces outcomes. Bluesky's profile in early 2026 — fast-growing, Western-backed, the primary destination for users migrating from Elon Musk's X — made it a sufficiently visible target without requiring a specific platform-level grievance. Security teams evaluating exposure to this class of actor should assess visibility and perceived alignment, not wait for an explicit threat to be stated before the risk is treated as real.
0 / 4 answered

CyberSpit Notes

  1. API servers are critical infrastructure, not supporting infrastructure. The Bluesky outage was not a network failure or a database failure. It was an API failure. Every user-facing function on the platform ran through the API layer, and when that layer was degraded, nothing worked. Security investment must reflect the functional centrality of APIs — they are the most impactful single point of failure for any service-oriented platform, and 313 Team's own Telegram statement explicitly named Bluesky's API as the operational target.
  2. Decentralization does not distribute the attack surface unless users actually distribute themselves. The AT Protocol is technically decentralized. The Bluesky user base is not. Third-party operators running their own AT Protocol infrastructure, like Blacksky, were unaffected; Bluesky's users, concentrated on Bluesky's servers, bore the full impact. Mastodon's federated instance model showed the inverse outcome: an attack on mastodon.social left the rest of the Fediverse untouched. Security posture must be evaluated against actual deployment patterns, not theoretical architecture.
  3. Attribution requires independent corroboration, but the 313 Team claim is consistent with the group's operational pattern. Bluesky correctly declined to speculate on attribution during an active incident. At the same time, 313 Team's Telegram claim, its targeting history against Western-aligned platforms (Microsoft, X, Cinemark, Amazon Prime Video, Yahoo, AOL, Clever, and Gulf government portals), and the follow-up Mastodon attack five days later all fit the group's established tempo. Security teams should monitor hacktivist claim channels as one input in a broader attribution picture, never as ground truth.
  4. DDoS should always be investigated as a potential cover event. Bluesky found no evidence of concurrent unauthorized data access, but that determination requires active investigation. Any sustained DDoS against an API-bearing platform should trigger a parallel review of authentication logs, access anomalies, and data pipeline activity during the degraded window — the same window during which monitoring capacity is most compromised.
  5. The broader API threat environment in 2026 makes this class of attack table-stakes planning. With Layer 7 DDoS up 104% over two years per Akamai, DDoS attacks surging 121% in 2025 per Cloudflare, and 87% of organizations reporting API security incidents, this is not a tail risk. Organizations that have not implemented multi-layer API DDoS resilience — edge mitigation, identity-aware rate limiting, endpoint inventory, isolated monitoring infrastructure — are not prepared for the current threat environment.
  6. AT Protocol's public Lexicon and IETF submission have made Bluesky's API surface formally catalogued attack reference material. The complete XRPC endpoint inventory, including parameter schemas, authentication requirements, and data types for every method, is published in the AT Protocol Lexicon repository and has been submitted to the IETF as Internet Drafts. This means that an attacker targeting Bluesky does not need recon tooling to identify compute-heavy endpoints: they can read the specification and identify app.bsky.feed.getTimeline, app.bsky.feed.getFeed, and similar aggregation calls as slow-path targets from the documentation alone. Defenders building threat models for open-protocol platforms need to account for this inversion: the same transparency that makes the platform trustworthy to developers makes its most expensive operations enumerable by adversaries before any reconnaissance occurs.
  7. The absence of a DDoS post-mortem is a transparency gap worth noting. When Bluesky's April 4-6 outage resolved, a detailed engineering post-mortem was published within four days. When the larger, externally driven DDoS incident resolved, no comparable write-up followed. This is not unusual — DDoS post-mortems involve attribution complexity, legal considerations, and competitive risk that internal engineering failures do not — but the absence means the public record for a significant security incident at a major platform contains confirmed facts about impact and almost nothing confirmed about defense. Security practitioners cannot learn what failed, in what order, or what worked from Bluesky's public communications alone.
  8. Target selection without stated motive is still traceable. 313 Team provided no specific justification for targeting Bluesky. Heise Online explicitly noted the group does not explain its motivations and merely reports outcomes. This pattern is consistent across the group's operations. Bluesky's profile in early 2026 — a fast-growing Western platform that had become the primary destination for users departing Elon Musk's X — was likely sufficient to meet the group's visibility threshold. Security teams evaluating exposure to 313 Team and similar actors should not wait for explicit grievances before treating the risk as real. High-visibility Western platforms with perceived alignment to US or Israeli geopolitical positioning are in scope for these groups whether or not a specific triggering event is named.

The Bluesky incident closed without a confirmed data breach and with attacker attribution that the platform itself has not publicly confirmed. In incident terms, that is a relatively contained outcome — even acknowledging that the attack activity continued intermittently through April 20 rather than cleanly ending after the first 24 hours. Five days into the incident, Mastodon's handling of a similar DDoS showed what it looks like when the federated model works as intended. But the attack exposed real fragility in the API infrastructure of a fast-growing platform at a moment when that infrastructure was under maximum load. For security practitioners, the incident is a high-fidelity case study in how API servers fail under pressure, who is likely to push them, and what it costs operationally when they do — with the honest caveat that much of the detailed forensic telemetry remains inside Bluesky, and public reporting on every dimension of this event carries real uncertainty.

Frequently Asked Questions

Pro-Iran hacktivist group 313 Team (also known as Islamic Cyber Resistance in Iraq) claimed responsibility via Telegram on April 16, 2026. The group has been active since June 2023 and operates within a coordinated Electronic Operations Room documented by SecurityScorecard and Unit 42. Bluesky itself declined to publicly attribute the attack — standard incident-response practice. The claim is consistent with the group's established operational pattern but has not been independently verified by a third party.

The concentrated disruption lasted approximately 24 hours. StatusGator external monitoring logged 17 hours and 17 minutes of downtime on April 16, plus an additional 1 hour and 31 minutes on April 17. Bluesky reported services stable by the evening of April 16, but intermittent attack activity continued through April 20, 2026, when a follow-up attack also struck mastodon.social.

Bluesky's API layer is the functional core of the entire platform. Every user interaction — loading a feed, fetching a profile, posting, receiving notifications — is mediated by API calls. Disrupting the API layer disrupts everything simultaneously, making it the highest-impact target per unit of attack traffic. Additionally, the AT Protocol publishes its full endpoint inventory in the Lexicon specification and as IETF Internet Drafts, meaning attackers can identify compute-heavy endpoints from documentation alone without active reconnaissance.

Bluesky found no evidence of unauthorized data access during or after the attack. However, that determination requires active investigation rather than assumption. Any sustained DDoS against an API platform should trigger a parallel review of authentication logs and data pipeline activity, because the degraded monitoring window is precisely when concurrent intrusion attempts are hardest to detect.

313 Team (Islamic Cyber Resistance in Iraq) is a pro-Iran hacktivist collective active since June 2023. The group is ideologically aligned with Iran's Wilayat al-Faqih framework and targets organizations associated with countries perceived as supporting the US or Israel. Its operations include DDoS attacks, website defacement, and Telegram-based psychological amplification campaigns. The name references the 313 warriors at the Battle of Badr in 624 CE. Security researchers place the group within a broader coordinated Electronic Operations Room alongside other Iran-linked hacktivist collectives.

AT Protocol is architecturally decentralized, but Bluesky's user base is operationally centralized. Over 43 million users rely on infrastructure operated by Bluesky the company — the primary Personal Data Server, the Big Graph Server, and the App View. Third-party AT Protocol operators such as Blacksky remained fully operational throughout the attack. The contrast with Mastodon is instructive: when mastodon.social was attacked on April 20, the broader Fediverse was unaffected because its users are genuinely distributed across thousands of independent instances.

A network-layer (Layer 3/4) DDoS overwhelms a target with raw packet or bandwidth volume, which scrubbing centers and upstream BGP controls can filter by source IP or traffic pattern. A Layer 7 (application-layer) DDoS sends structurally valid HTTP requests to application endpoints — requests that look identical to legitimate user traffic at the network level. Against an API platform like Bluesky, attackers target the most compute-expensive endpoints, such as full-graph feed aggregation calls, so each malicious request consumes disproportionate server resources. Standard IP-based rate limiting is insufficient because modern Layer 7 floods distribute requests across large botnet pools, keeping each source below individual thresholds while generating destructive aggregate load. Effective defense requires identity-aware throttling, endpoint-specific limits tuned to compute cost, and monitoring infrastructure isolated from the production systems under attack.

Sources and Further Reading

  • Bluesky official status posts on @bsky.app and @status.bsky.app, April 16–20, 2026
  • Bluesky — "Bluesky Service Interruption" blog post (bsky.social/about/blog/04-16-2026-bluesky-service-interruption)
  • Bluesky — "Bluesky Service Interruption Update" blog post series (April 20, 2026 update-three)
  • Bluesky — "A New Chapter for Bluesky" leadership announcement (March 9, 2026)
  • Jim Calabro — "April 2026 Outage Post-Mortem" engineering writeup on pckt.blog (April 10, 2026) — primary source for the April 4-6 port exhaustion incident
  • Toni Schneider — "Coming Off the Bench for Bluesky" personal blog post (toni.org, March 9, 2026)
  • TechCrunch — "Bluesky confirms DDoS attack is cause of continued app outages" (April 17, 2026)
  • TechCrunch — "Mastodon says its flagship server was hit by a DDoS attack" (April 20, 2026)
  • The Record / Recorded Future News — "Bluesky blames app outage on 'sophisticated' DDoS attack" and "Russia appears to block social media platform Bluesky amid wider internet restrictions"
  • SecurityWeek — "Bluesky Disrupted by Sophisticated DDoS Attack" by Eduard Kovacs (April 20, 2026)
  • Security Affairs — "Bluesky hit by 24-hour DDoS attack as pro-Iran group claims responsibility" by Pierluigi Paganini
  • Security Affairs — "DDoS wave continues as Mastodon hit after Bluesky incident"
  • Hackread — "Bluesky Back Online After DDoS Attack, as Iran-Linked 313 Team Takes Credit" by Deeba Ahmed (April 22, 2026)
  • Heise Online — "Bluesky outage: DDoS attack allegedly by Iranian group 313 Team"
  • Threat Beat — "Pro-Iran hackers claim 'sophisticated' attack on Bluesky" — primary source for the 313 Team Telegram quote
  • Engadget — "Bluesky blames DDoS attack for server outages" by Kris Holt (April 16, 2026)
  • Engadget — "Mastodon was hit by a 'major' DDoS attack" (April 20, 2026)
  • Mashable — "Bluesky breaks silence on outage and reveals cause" (April 2026)
  • StatusGator — Bluesky outage duration telemetry (statusgator.com/services/bluesky)
  • Akamai — 2026 Apps, APIs, and DDoS State of the Internet Report (March 17, 2026)
  • Cloudflare — 2025 Q4 DDoS Threat Report (February 2026)
  • HawkEye Threat Advisory — 313 Team / Islamic Cyber Resistance profile
  • Daily Dark Web — "Exclusive Interview: 313 Team" (April 2026) — primary source for group's stated capability trajectory and ideological framing
  • Unit 42 (Palo Alto Networks) — "Threat Brief: Escalation of Cyber Risk Related to Iran" (Updated April 17, 2026) — primary source for Electronic Operations Room and IRGC coordination context
  • SecurityScorecard STRIKE Team — analysis of 250,000 messages from 178 active Iran-aligned hacktivist groups, as cited in Unit 42 and GNET Research
  • GNET Research — "Al-Qaeda's Cyber Jihad Movement: Plugging into Iran's Wartime Hacktivist Ecosystem" (March 2026)
  • Surfing Complexity (Lorin Hochstein) — "Thoughts on the Bluesky public incident write-up" technical analysis (April 12, 2026)
  • Wikipedia — Bluesky and Jay Graber articles (for leadership history and corporate structure)
  • CNBC — "Bluesky CEO Graber stepping back, Toni Schneider named interim boss" (March 9, 2026)
  • NIST SP 800-228 — Guidelines for API Protection for Cloud-Native Systems, Chandramouli and Butcher (June 2025) — finalized NIST guidance on API lifecycle risk, rate limiting, API inventory, and protection controls: csrc.nist.gov/pubs/sp/800/228/final
  • NIST SP 800-189 — Resilient Interdomain Traffic Exchange: BGP Security and DDoS Mitigation (final); Rev. 1 initial public draft released January 2025 — upstream DDoS mitigation including RTBH filtering, Flowspec, uRPF, and SAV: csrc.nist.gov/pubs/sp/800/189/final
  • NIST SP 800-61r3 — Incident Response Recommendations and Considerations for Cybersecurity Risk Management: A CSF 2.0 Community Profile, Nelson, Rekhi, Souppaya, Scarfone (2025) — supersedes SP 800-61r2: csrc.nist.gov/pubs/sp/800/61/r3/final
  • NIST SP 800-53 Rev. 5 — SC-5 (Denial-of-Service Protection) and SI-10 (Information Input Validation) — baseline security controls applicable to API DDoS resilience and schema validation at the gateway layer
  • MITRE ATT&CK Enterprise — T1498 (Network Denial of Service), T1498.001 (Direct Network Flood), T1499 (Endpoint Denial of Service), T1499.002 (Service Exhaustion Flood), T1499.003 (Application Exhaustion Flood), T1583.005 (Acquire Infrastructure: Botnet), T1590.002 (Gather Victim Network Information): attack.mitre.org
  • AT Protocol specification — HTTP API (XRPC) technical specification (atproto.com/specs/xrpc) — primary source for XRPC endpoint structure, NSID addressing, and authentication scheme
  • Bluesky Engineering Blog — AT Protocol federation architecture documentation (docs.bsky.app) — primary source for PDS, BGS, and App View layer descriptions
  • @atproto/xrpc-server library documentation — primary source for Redis rate limiter implementation and rate limit configuration (mintlify.com/bluesky-social/atproto/api/server/xrpc-server)
  • Wikipedia — AT Protocol article (August 2025 IETF BOF submission documentation)
  • HawkEye Threat Advisory — 313 Team SIGMA detection rules and CheckHost.net proof methodology documentation (March 11, 2026)
  • Unit 42 (Palo Alto Networks) — CL-STA-1128 / Cyber Av3ngers Rockwell Automation FactoryTalk OT/ICS targeting documentation (April 2026 threat brief update)