Anthropic's Claude API Took a 38-Minute Hit on Sep 29 - Claude.ai, Claude Code, Claude Cowork, and the API Were All Affected. Some Messages May Have Been Lost.

Anthropic's status page records Incident 31554916 on Sep 29: 38 minutes of elevated errors from 14:00-14:59 UTC affecting Claude.ai, Claude Code, Claude Cowork, the Claude API, and platform.claude.com. A follow-on SSO/sign-in tail ran through ~15:30 UTC. Status page is candid: 'Some messages sent between 14:00 and 14:59 UTC may not have been saved.' Second Claude API incident in seven days.

Originally published: 2026-09-29 20:08 UTC / 22:08 Berlin / 16:08 EDT

What Happened

Anthropic's status page (status.claude.com) records a single incident on Sep 29, 2026, titled "Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API" (Incident ID 31554916). The incident timeline:

14:21 UTC - Investigating elevated error rates affecting Claude.ai (desktop and mobile), Claude Code, and Claude Cowork. "Users may see failed requests, errors loading or sending conversations, or be asked to sign in again; retrying may succeed."
14:28 UTC - Confirmed scope expands: "Some requests to the Claude API are also returning errors."
14:30 UTC - Confirmed platform.claude.com and the Claude API itself are also degraded.
14:36 UTC - Mitigation applied; "error rates have dropped substantially." Some sign-in attempts, account actions, and Claude Code/Cowork sessions still failing.
14:41 UTC - Degraded performance continues across claude.ai, platform.claude.com, Claude API, Claude Code, Claude Cowork.
14:59 UTC - Most services recovered. A second issue appears: "single sign-on and Sign in with Apple are unavailable," plus new chats, voice conversations, Claude Code/Cowork sessions, purchases, and file uploads are blocked for affected users.
15:11 UTC - Most services recovered as of 14:59 UTC. "Some messages sent between 14:00 and 14:59 UTC may not have been saved."
15:30 UTC - All services operating normally since 14:59 UTC.
16:27 UTC - Final resolved post. Impact window: 14:00 UTC to 14:59 UTC, 38 minutes of elevated errors. Then a secondary SSO/sign-in tail extending through ~15:30 UTC.

This is the second Claude API elevated-error incident inside seven days. The Sep 22 incident hit Opus 5, Fable 5/5.1, and Mythos 5 with elevated errors for ~1h20m. Today's hit is shorter (38 min) but broader in product scope.

What Actually Changed

For users hitting Claude products between 14:00 and 14:59 UTC on Sep 29:

Claude.ai, Claude desktop, Claude mobile apps: requests failed or errored out; conversations failed to load or send; users were asked to sign in again.
Claude Code: sessions failed or stalled; existing conversations affected.
Claude Cowork: sessions affected.
Claude API (the developer-facing endpoint at api.anthropic.com): "Some requests to the Claude API are also returning errors." Builder workloads running automated agents, batch jobs, or interactive Claude Code sessions during this window saw failures.
platform.claude.com (console): degraded performance.

For users trying to start new work between ~14:59 and 15:30 UTC:

Single sign-on and "Sign in with Apple" unavailable.
New chats, voice conversations, Claude Code and Cowork sessions, purchases, file uploads all blocked for affected users.
Existing signed-in users were advised not to sign out.

Data-loss implication (15:11 UTC update): "Some messages sent between 14:00 and 14:59 UTC may not have been saved." This is a documented message-persistence warning from the operator, not an inferred risk.

Why Developers and Founders Should Care

If you run Claude Code in production - long-running agents, autonomous workers, scheduled batch jobs - the 14:00-14:59 UTC window is when your runs could have failed. Three categories of impact:

1. Failed API requests during the window. Pull your access logs for api.anthropic.com calls between 14:00 and 14:59 UTC. Retry-after, 5xx, connection-reset, and timeout errors are the patterns. If you don't have retries on every Claude API call, you're missing data on this incident. 2. Lost messages. The 15:11 UTC update says some messages sent in the 14:00-14:59 UTC window may not have been saved. If you use Claude.ai as a workspace, audit conversations that were active in that window for missing turns. 3. SSO/sign-in disruption tail. If your team uses SSO to log into Claude.ai, Claude Code, or Cowork, expect some sign-in failures between ~14:59 and ~15:30 UTC. Anyone who signed out during the incident lost access for the tail; anyone who stayed signed in kept working.

For the broader reliability story: this is the second Claude API incident in seven days. Anthropic does not publish incident-class SLAs publicly. The repeated pattern - broad product surface, ~38 minutes to ~1h20m, mid-afternoon US East - is worth factoring into your runbook assumptions about Claude availability for latency-sensitive workflows.

Evidence and Test Results

All claims above are from the status page incident timeline:

Incident 31554916 - Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API (final resolved post at 16:27 UTC).
Status page incident history Atom feed (verbatim timestamps and update text).

Specific verbatim claims from the status page:

14:21 UTC investigating: "Users may see failed requests, errors loading or sending conversations, or be asked to sign in again; retrying may succeed."
14:30 UTC update: "We are investigating reports of degraded performance affecting claude.ai, platform.claude.com, the Claude API, Claude Code, and Claude Cowork."
15:00 UTC update: "The errors that began at 14:00 UTC were mitigated at 14:36 UTC, and existing conversations are mostly working again. A second issue is preventing many users from signing in (single sign-on and Sign in with Apple are unavailable), starting new chats, voice conversations and Claude Code or Cowork sessions, making purchases, and uploading files."
15:11 UTC monitoring: "Some messages sent between 14:00 and 14:59 UTC may not have been saved."
16:27 UTC resolved: "The issue affecting Claude.ai, the Claude desktop and mobile apps, Claude Code, Claude Cowork, and the Claude API has been resolved, and services have been operating normally since recovery. Impact occurred from 07:00 PT / 14:00 UTC to 07:59 PT / 14:59 UTC."

Verification level: documentation comparison against the live status page and Atom feed. No independent probe of api.anthropic.com was run for this report. The 38-minute window and the message-loss advisory are both operator-published; treat them as authoritative.

Cost, Risk, and Limitations

Operational cost. If your Claude Code or API workflow was active in the 14:00-14:59 UTC window, expect partial or full loss of the work that was in flight. Token spend during failed requests may or may not have been billed depending on where in the request lifecycle the failure occurred; the status page does not quantify. Pull your usage export to confirm.
Data loss scope is unclear. The advisory says "some messages ... may not have been saved" without quantifying how many or which. If you depend on Claude.ai chat history as a system of record, treat the 14:00-14:59 UTC window as a recovery window - re-derive or re-send any work product you need.
No public root-cause. The Sep 29 resolved post does not publish a root-cause analysis. The Sep 22 post (Opus/Fable/Mythos elevated errors) also did not publish a root-cause. Two incidents in seven days without published post-mortems is a signal worth filing under "vendor reliability evidence," not "vendor unreliability panic."
No SLA. Anthropic does not publish a public uptime SLA or incident-class commitment for Claude.ai or the Claude API. If your SLOs require a specific monthly error budget, you're computing against an absence of published numbers.
The pattern is repeatable. 38 min today, 1h20m on Sep 22, prior incidents earlier in the year. If you run a multi-model agent stack, Claude should be treated as "good but occasionally unavailable for ~30-90 min" rather than "always on."

Mr. Technology Verdict

A 38-minute elevated-error incident affecting the full Claude product surface - Claude.ai, Claude Code, Claude Cowork, the Claude API, and the platform console - is a notable reliability event even if it doesn't rise to the level of a multi-hour outage. The status page post is candid: it documents the failure window, the message-loss advisory, and the SSO/sign-in tail. Builders using Claude as a primary agent backbone should treat the Sep 22 + Sep 29 pair as evidence that a single-vendor Claude-only architecture carries concentrated vendor-reliability risk.

The change for builders is not "stop using Claude." It is "make sure your Claude integration has retries with jitter, an error budget for ~30-90 min monthly downtime, and a fallback model for the latency-sensitive path." The status page is not promising otherwise.

Recommended Action

1. Today: Pull api.anthropic.com access logs for 14:00-14:59 UTC. Tally 5xx, 429, retry-after, connection-reset, and timeout errors. If your error rate during this window is non-zero, confirm your retry policy absorbed the failures and surface the loss to anyone tracking SLOs. 2. Today: If anyone on your team was using Claude.ai or Cowork for work product between 14:00 and 14:59 UTC, have them verify the affected conversations for missing turns. The status page is explicit that some messages may not have been saved. 3. This week: If you don't already have a fallback model wired into your agent backbone, add one. A second-model failover path (Anthropic Sonnet 5 -> OpenAI GPT-6 family, or whatever fits your stack) is the single highest-leverage change for absorbing this class of incident. 4. This week: Audit your Claude Code retry policy. Default Codex/Claude-Code retry-on-5xx behavior is reasonable, but verify that retry-after is honored, jitter is applied, and idempotency keys are set for any non-GET API calls. 5. Skip if not in scope: If you don't run Claude Code or the Claude API in production, this incident does not affect your stack today.

Sources

Last verified: 2026-09-29 20:08 UTC. No corrections on file.