Anthropic's spend-limit check wrongly paused paying organisations on October 7
Anthropic's own spend-limit check refused requests from paying customers on October 7, after deciding they had hit a cap they had not hit. The company marked the fault resolved in an update posted at 21:23 UTC that day on its status page: "We have identified an issue which resulted in some organizations to be incorrectly paused for having reached their spend limit, resulting in refused requests on the Claude API, Claude.ai, Claude Code, Claude Security, and Claude Cowork."
An Anthropic illustration of an easel and a drawn scribble, on orange. Image: Anthropic
Five products failed for one reason. A spend limit is held against an organisation rather than against a product, so a developer calling the API, somebody typing into Claude.ai and an agent running in Claude Code all met the same refusal.
That cap is the monthly ceiling on an organisation's Claude API spending, and Anthropic's documentation sets it at $500 on the Start tier, $1,000 on Build and $200,000 on Scale, in US dollars, with no cap on the Custom tier. An organisation that reaches its cap has its API usage paused, and requests come back as an HTTP 429 carrying the error code enforced_spend_limit_reached. That is the state some paying organisations were put into without having spent the money.
Anthropic published no window for it
The entry carries one update and no start time. Anthropic usually gives both ends when it has them. An earlier incident on the same page, covering errors loading usage data on platform.claude.com, is dated "from 17:00 PT on October 6 / 00:00 UTC on October 7 to 18:40 PT on October 6 / 01:40 UTC on October 7", and the errors affecting Claude Mythos 5.1 and Claude Fable 5.1 on October 5 are given as 05:40 PT / 12:40 UTC to 06:10 PT / 13:10 UTC. For the spend-limit fault there is a resolution stamp and nothing before it.
That usage-data incident also shows what a running update from Anthropic looks like. At 01:20 UTC on October 7 the company wrote that it was investigating the errors "since 5 PM PT" and added that "Messages API requests are not affected", scoping the damage while the fault was still open. The resolution followed at 02:33 UTC. The spend-limit entry appeared only once it was over.
Scope is equally thin. "Some organizations" is the whole of it, with no count, no region and no plan named. Anthropic describes refused requests rather than incorrect charges, and says nothing about work lost while accounts were paused.
A second billing fault hit the status page in October
October has two of them. On October 1 the same status page logged delayed credits on the Claude Platform, with Anthropic investigating from 16:20 UTC, moving to monitoring at 18:23 and resolving at 22:36 after it "resolved the issue causing delays in credits being applied after purchase". That one also ended in requests failing, for insufficient credit balance rather than a spend cap. They are two separate entries, and Anthropic has not linked them to a common cause.
Anthropic has been changing how Claude requests are billed as well as how they fail. The company began charging for refused Claude API requests in three of five categories in September.
Model faults on the same page carry timings at both ends. Elevated errors for Claude Opus 5.5 ran from an investigating note at 12:24 UTC on October 6 to a resolution at 12:43, and the September 29 outage, after which Anthropic said some messages sent during it may not have been saved, was given an exact impact window.
The status page was read at 21:04 UTC on October 8. Anyone whose requests failed on October 7 has Anthropic's answer, which is that the refusal came from a spend limit the account had not actually reached.
