Reading
Claude Lost Your Messages? Anthropic Still Won't Say Why
Anthropic says some Claude messages from a one-hour outage on Sept 29 may not have been saved. It gave no count and no cause. Here's what to check.
Prices last read 2026-10-07
Short answer
If you used Claude between 14:00 and 14:59 UTC on September 29, 2026, some of what you sent may not have been saved. Anthropic hasn't said why.Claude's official status page says the incident affected claude.ai (web, desktop and mobile), Claude Code, Claude Cowork, the Claude API and the Console. It says 'some messages sent between 14:00 and 14:59 UTC may not have been saved.' It posted eight updates and then marked the incident resolved at 16:27 UTC. The incident report gives no root cause and no estimate of how many messages were affected. As of today, we couldn't find a post-mortem for it.Source: Claude Status: Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API · 2026-10-07
What actually happened: two failures in one hour
Anthropic's own timeline shows two separate problems. Errors started at 14:00 UTC, and the first status update went up at 14:21. By 14:30 the incident covered claude.ai, the API and platform.claude.com. A mitigation update followed at 14:41. Then, at 15:00, Anthropic reported a second issue: single sign-on and Sign in with Apple were down, and so were new chats, voice conversations, new Claude Code and Cowork sessions, purchases and file uploads. Anthropic says most services had recovered by 14:59 UTC. The 15:11 update added the warning about unsaved messages.
Source: Claude Status: Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API · 2026-10-07
According to The Decoder (Mixed News), Anthropic did not confirm whether any data was actually lost. It also didn't say how many messages were affected or whether those messages reached Claude's servers. The same report says Anthropic gave signed-in users this advice during the sign-in problem: 'If you're signed in, please don't sign out.' 9to5Google reports that this was Claude's fourth outage in September, after incidents on September 11, 15 and 22.
Source: Mixed News: Anthropic's Claude September 29 outage – messages may not have been saved · 2026-10-07; 9to5Google: Claude confirmed outage, Sept 29 · 2026-10-07
Who should check what
- Chat users (web, desktop, mobile)Open every conversation you touched between 14:00 and 14:59 UTC on September 29. That's 7:00–7:59 a.m. Pacific. Look for prompts or replies that aren't there. If something important is gone, rebuild it now while you still remember it.
- Claude Code and Cowork usersDon't assume the session history is complete. Check your repository's actual state with git status and git log. Don't rely on the agent's memory of what it changed. Re-run any task that was in progress during that hour.
- API developersCheck your own logs for requests in that hour that returned errors or timed out, and replay the ones that matter. Your logs are the only record you control.
- EveryoneYou can subscribe to updates on status.claude.com. If you lost work you can't rebuild, contact Anthropic support. Mention the incident window, and ask directly whether your messages were stored.
THE SERVICES AND OUTPUTS ARE PROVIDED "AS IS" AND "AS AVAILABLE" WITHOUT WARRANTY OF ANY KIND
Anthropic Commercial Terms of Service (effective June 17, 2025)Source: Anthropic Commercial Terms of Service · 2026-10-07
This clause comes from Anthropic's Commercial Terms, which cover API and business use. The same terms also say Anthropic does not warrant that use of the services will be uninterrupted. Anthropic's Consumer Terms (effective October 8, 2025) contain the same 'as is' and 'as available' disclaimer. They also let Anthropic modify, suspend or discontinue the services without notice. In practice, nothing in the contract obliges Anthropic to publish a post-mortem or to say how many messages were lost.
Source: Anthropic Commercial Terms of Service · 2026-10-07; Anthropic Consumer Terms of Service · 2026-10-07
Our take: "may not have been saved" isn't an answer
Outages happen, and we're not going to pretend any AI company is immune. The outage itself isn't what bothers us. A paid service told customers their data might be gone and then went quiet. 'May not have been saved' is the most annoying phrasing possible. It's vague enough to avoid admitting anything, and alarming enough that every heavy user now has to audit their own history. Was it one message or a million? Did the model ever get the prompt? Was it a database write failing, or an authentication cascade? Anthropic hasn't said.
This hits Claude harder than a typical chat app. Claude Code and Cowork run agents that change real files. If an agent's session record and the actual state of your repository disagree, that's a working problem, not just a cosmetic one. A company that sells agents to developers should know that a timeline of eight status updates is not an explanation.
Our position is simple. Losing data deserves a written post-mortem. Pro and Max subscribers, and API customers who pay by the token, should get a root cause for both failures, a rough scale of the message loss, and a clear statement on whether the lost messages can be recovered. Anthropic publishes long essays about AI safety and transparency. The least it can do is apply some of that transparency to an hour when the safety of its customers' own data was at stake. Until it does, treat your own logs and git history as the record of what happened, not Claude's chat history.
Habit worth keeping
Once a conversation becomes something you'd hate to lose, copy the key output somewhere you control, whether that's a doc, a repo or a notes app. Do this no matter which AI service you use.
Optional: compare Claude subscription listings on SlotPrices. These are seller listings, not official Anthropic access and not an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.
Sources (all read on 2026-10-07)