Claude outage – Claude’s platform suffered a tenth significant disruption since June 5, with errors affecting Claude Opus 4.8 and Haiku 4.5 that began Tuesday afternoon. The incident was declared fixed, then errors persisted hours later as users scrambled to confirm whether t
On Tuesday afternoon, the errors started at 1:29 PM ET—first hitting Claude Opus 4.8 and Haiku 4.5. By late afternoon, Anthropic said it had a fix in place. Then, at 6:15 PM UTC, the company’s own status updates reported the same problem was still there.
The outage is now the tenth significant service disruption since June 5. As of the most recent status update, the incident remained unresolved, and it continued to affect users of claude.ai, the Claude API, Claude Code, and Claude Cowork.
Anthropic’s status page put the timeline in plain terms. It confirmed the incident began at 5:29 PM UTC (1:29 PM ET) on Tuesday. A fix was declared at 6:00 PM UTC. At 6:15 PM UTC, a follow-up update said errors on Opus 4.8 and Haiku 4.5 persisted. By 7:00 PM UTC (3:00 PM ET). the company said it was “continuing to work to resolve an elevated rate of errors on requests to Claude Opus 4.8. ” and promised another update “as soon as possible.”.
Search traffic for “is Claude down” spiked on Google Trends as users tried to verify what they were seeing.
Claude Console and Claude for Government appeared unaffected by the current incident.
The silence around what went wrong has become part of the story. Anthropic did not issue a standalone public statement about the June 16 incident beyond its rolling status page updates. That continued an earlier pattern: a review of public reporting found the company has not published post-incident root cause analyses or detailed engineering post-mortems for any of its recent June disruptions.
In April 2026, Anthropic did offer a more direct explanation—just not in a separate breakdown tied to each outage. In a statement to Fortune. the company said “Demand for Claude has grown at an unprecedented rate. and our infrastructure has been stretched to meet it. particularly at peak hours.” It also said “Compute is a constraint across the entire industry. ” adding that it is scaling compute “rapidly and responsibly. ” including through an expansion of its partnership with Amazon and Google that it said would bring “significant new capacity online in the coming months.”.
Those additional capacity plans have not arrived yet, and the disruptions have kept coming.
This outage lands the day after a proposed class-action lawsuit accused Anthropic of delivering far less AI service than premium subscribers paid for.
The lawsuit was filed June 14, 2026, in the Northern District of California by plaintiff Karl Kahn. The complaint alleges that Anthropic’s Claude Max plans deliver far less usage than advertised: the Max 20x plan is said to provide only six to eight times the base Pro tier’s usage rather than the promised twenty times. Anthropic declined to comment.
On the service reliability side, Tuesday’s disruption fits into a steep run of problems documented since June 5.
Independent monitoring by StatusGator confirmed the following incidents for claude.ai since that date:
June 5 — Significant outage affecting claude.ai, the API, Claude Code, and Claude Cowork. Anthropic investigated user reports of potential data leakage and found no evidence of customer data exposure, attributing the disruption to an infrastructure failure.
June 7 — Warning-level incident lasting 1 hour and 19 minutes.
June 8 — Warning-level incident lasting 2 hours and 4 minutes.
June 9 — Two separate incidents: a 25-minute warning and a 1-hour-40-minute degradation.
June 10 — Warning-level incident lasting 3 hours and 14 minutes.
June 11 — Two separate incidents: a 24-minute warning and a 29-minute outage.
June 13 — Warning-level incident lasting 39 minutes.
June 15 — Warning-level incident lasting 1 hour and 54 minutes.
Taken together, that sequence puts the mean time between significant failures at roughly one day—something that would breach standard commercial service-level agreements that enterprise cloud providers typically guarantee at 99.9% uptime or higher.
For the people most exposed, outages aren’t just an inconvenience. Claude Code is increasingly deployed inside automated workflows. A February 2026 SemiAnalysis report said Claude Code accounts for roughly 4% of all public GitHub commits, processing more than 135,000 commits per day.
Unlike a simple chatbot session—where a user can wait and retry—Claude Code is often embedded in CI/CD pipelines that execute multi-step tasks autonomously. When an outage interrupts a task mid-run, the pipeline does not pause and resume. Instead, it fails and can require manual intervention to clean up partially executed operations.
Thoughtworks, in remarks after the June 2 outage, pointed to single-vendor dependency as the threat to business continuity. It said that hardcoding a specific provider’s API endpoint into an application was an acceptable availability strategy in the early days of AI adoption. but that “in 2026. it’s a single point of failure that’s a very real threat to business continuity.”.
During the June disruptions, developers have also had to contend with different kinds of errors. Anthropic distinguishes between HTTP 500 errors—indicating an internal server failure inside Anthropic’s infrastructure—and HTTP 529 errors. which signal capacity overload. Anthropic’s API documentation says an HTTP 529 error “does not count against your quota” and resolves when capacity frees up. while a 500 error requires a server-side fix.
Tuesday’s incident remains tangled up in that same capacity story.
The strain is tied to multiple factors that converged over the past year. Runtime.news reported in March 2026 that Anthropic’s infrastructure has struggled to keep pace with the surge in demand for its Opus and Sonnet models following Claude Code’s rapid growth. That growth. the report said. was compounded by a 55% jump in Claude mobile app downloads after Anthropic’s public dispute with the Pentagon over AI safety restrictions.
In that dispute, Anthropic refused to allow Claude’s use for fully autonomous weapons or mass domestic surveillance. Afterward, the Trump administration designated the company a supply-chain risk, and the report said that designation drove users away from competitors and toward Claude.
The customer numbers behind the infrastructure pressure are stark. Anthropic’s annualized revenue climbed from $9 billion at the end of 2025 to more than $30 billion in early April 2026. Over the same period. the number of enterprise clients spending more than $1 million annually doubled from 500 to 1. 000 in under two months.
That growth came faster than the infrastructure was built to handle.
For the people trying to operate through the outage right now, there are still specific steps available. Users and developers can monitor live updates at status.claude.com. The page breaks down the health of individual services—claude.ai. the API. Claude Code. and Claude Cowork—and supports email. SMS. Slack. and Microsoft Teams notifications for incident alerts.
For developers, Anthropic’s API documentation recommends switching to a different model tier during model-specific capacity events. The error message from Claude Code during overloads prompts users to run /model and select a less-constrained model.
As of Tuesday at 3:00 PM ET, the incident remained active.
And for anyone wondering whether the outage was happening “right now” on June 16, 2026, the answer at that point was yes. Anthropic’s status page confirmed an active incident affecting Claude Opus 4.8 and Haiku 4.5. with elevated error rates reported across claude.ai. the Claude API. Claude Code. and Claude Cowork. The company had attempted a fix at 2:00 PM ET. but said it failed to hold. and that it was continuing to work on resolution.
If Claude is unavailable, the guidance is practical: check status.claude.com, and subscribe to email, SMS, or Slack alerts. Developers using the Claude API should also note that the direct Anthropic API and AWS Bedrock run on separate infrastructure pools—during some outages. Bedrock remains available when the direct API does not.
For Claude Code users, running the /model command can switch to a different model tier when one model is experiencing capacity constraints.
The lawsuit and the outage pattern trace back to the same mismatch—between usage and what the infrastructure was designed to sustain. In the proposed class-action, the dispute is over what customers were promised in their subscription. In the outage logs, the dispute is over what the systems can actually deliver when demand spikes.
Right now, the platform’s message is still the same one that’s been repeated through each status update: errors elevated, a fix attempted, and the system still not fully back to normal as the hours pass.
Claude outage Anthropic Opus 4.8 Haiku 4.5 Claude Code Claude API service disruption status.claude.com Karl Kahn lawsuit
So it’s down again??
I don’t get why they keep saying “fixed” and then it’s not fixed. Also 10 disruptions since June 5 is kinda wild. Prob just traffic from all the people using it, idk.
I saw this and thought it was just Haiku having issues but now it’s Opus too?? Like which one is the real brain lol. Every time I try Claude Code it acts like my account is “loading” forever. Hopefully they reimburse people if it costs dev time.
Man I hate when these AI tools go down, it’s always the same thing. They say error at like 1:29 PM ET then “fix” at 6:00 UTC like that somehow syncs with life. Also doesn’t “status updates” mean it should be already monitored? Sounds like they’re just tweeting updates after the fact. If it’s affecting claude.ai and the API and everything, that’s basically the whole product, not just some minor hiccup.