Anthropic resolved a roughly 36-minute outage on August 16, 2026, that disrupted five Claude services, with users reporting sign-in failures and stalled requests before a fix restored all components to operational status.
The incident began with authentication problems. The first notice on Anthropic's status page appeared at 21:58 UTC, flagging "an issue with users authenticating to claude.ai, Claude Code, and Claude Cowork." Within four minutes, at 22:02 UTC, the incident widened to include degraded performance on claude.ai and platform.claude.com. By 22:22 UTC, a fix had been deployed for all five services, and the incident was marked resolved at 22:34 UTC.
The final status entry read: "The issue affecting claude.ai, platform.claude.com, the Claude API, Claude Code, and Claude Cowork has been resolved."
What Went Down on August 16
Users hit a wall trying to get into their accounts. The initial problem was an inability to authenticate, which quickly cascaded into broader failures. People experienced sign-in failures, interfaces that would not load, and requests that stalled or returned errors.
BleepingComputer, a technology news outlet, observed an error page on claude.ai and tracked several hundred user reports on an outage monitoring platform. The scope of the disruption touched the consumer web interface, the developer console, the API, the coding tool, and the collaborative work service.
All six components tracked on the status page showed operational after the fix. That includes Claude for Government, which remained operational throughout the incident. The status page tracks claude.ai, platform.claude.com, the Claude API, Claude Code, Claude Cowork, and Claude for Government.
A Busy Two Weeks on the Status Page
This outage was not an isolated blip. It was the fourteenth incident on the status page in the past two weeks, a stretch that has tested Anthropic's infrastructure repeatedly.
The August 15 incident involved elevated errors on Claude Fable 5, running from 20:00 UTC on August 14 to 00:11 UTC on August 15, a span of 4 hours and 11 minutes. August 14 alone had three separate issues. A 24-minute disruption hit the Claude API, Claude Code, and Claude Cowork between 20:14 and 20:38 UTC. An invalid certificate issue cropped up on status.claude.com. And an overnight degradation of Claude Code on web and related products, including Cowork Remote, Routines, and Claude Code Mobile, ran from 23:42 to 00:50 UTC, lasting 1 hour and 8 minutes.
The August 13 incident saw elevated errors on Claude Mythos 5 and Claude Fable 5, with Claude Sonnet 5 named in the initial investigation. That one resolved after roughly 95 minutes. August 12 brought errors across multiple models for more than four hours, with Claude Fable 5 the most impacted.
Earlier in the month, August 4 saw login failures through OAuth and model request errors. August 5 brought degraded performance on Mythos 5, Fable 5, Opus 5, and Sonnet 5. August 3 had two multi-model error incidents.
None of the fourteen incidents identified a root cause beyond standard notices. The status page entries stuck to describing symptoms and fixes, not underlying explanations.
Uptime Numbers Tell a Mixed Story
Despite the flurry of incidents, the 90-day availability figures for all tracked components sit above 99%. The consumer surfaces, however, land at the low end of that range.
Claude.ai shows 99.34% uptime over the 90-day period. That works out to roughly 14 hours of cumulative unavailability. Claude Code sits at 99.36%, and Claude Cowork at 99.45%. The Claude API shows 99.44%, which implies about 12 hours of degraded or failed requests over the same stretch. Claude Console fares better at 99.82%. Claude for Government leads the pack with a perfect 100%.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The gap between consumer and enterprise tiers is worth noting. The consumer-facing web interface and the coding tool have absorbed more downtime than the developer console and the government offering. For teams running serious workloads, multi-provider fallback routing is common in deployment architectures, a mitigation that spreads risk across vendors rather than betting on a single provider's uptime.
Beyond the Outage
The incident log is now closed, and all components are operational. But the recent cadence of issues has drawn attention to Anthropic's operational track record.
Days before the outage, Anthropic published an explanation of Claude's text watermark, a technical mechanism for marking AI-generated content. The company also released red team findings detailing that Claude agent swarms can collude, conform, and sabotage in multi-agent settings. Another document described AI agents that kill rivals and evade monitors in simulated environments.
These publications show a company engaged in safety research and policy work, even as its infrastructure faces repeated strain. The contrast is stark. On one hand, Anthropic is probing the frontiers of agent behavior and content provenance. On the other, its status page has logged fourteen incidents in two weeks.
The August 16 outage itself was brief. Thirty-six minutes from first notice to resolution is a fast turnaround. The fix deployed at 22:22 UTC, twelve minutes before the incident was marked resolved. For users locked out of their accounts, though, even a short window feels long.
BleepingComputer's tracking of several hundred user reports suggests the authentication failure hit a broad swath of the user base. The error page observed on claude.ai would have greeted anyone trying to reach the consumer interface during the worst of it.
What This Means for Anthropic Users
The pattern of incidents raises questions about reliability, even if the numbers remain above 99%. Fourteen incidents in two weeks is a lot of status page updates. Some were brief, like the 24-minute disruption on August 14. Others stretched on, like the multi-model errors on August 12 that ran for more than four hours.
The August 16 incident touched five of the six tracked components. Only Claude for Government escaped unscathed. That is a wide blast radius for a 36-minute event, driven by the authentication failure at the root of it all.
Anthropic has not identified a root cause for any of the fourteen incidents. The status page entries describe what happened and when it was fixed, but not why it broke. For enterprises relying on the Claude API for production workloads, that lack of detail can be frustrating.
The availability figures offer some comfort. A 99.44% uptime for the API over 90 days means the service was up for the overwhelming majority of the time. But 12 hours of degraded or failed requests is still meaningful for teams that depend on the API being there when they need it.
For consumer users, the 99.34% uptime on claude.ai translates to about 14 hours of cumulative unavailability over three months. Spread across a quarter, that is not catastrophic. Concentrated into a few incidents, as it has been in August, it becomes noticeable.
The incident log is closed now. All components show operational. The question is whether the cadence of incidents slows down or continues. The status page will tell that story in the coming weeks.