- Claude experienced five separate service disruptions in March 2026, including a nearly five-hour outage on March 28 — the longest in the platform’s history.
- Affected services included Claude.ai, Claude Code, Claude Cowork, the Developer Console, and MCP calls, though the core API remained functional during some incidents.
- The outages coincided with a visible surge of users migrating from ChatGPT to Claude following OpenAI’s Pentagon deal announcement, though Anthropic has not confirmed this as the root cause.
What Happened
Anthropic’s Claude went down five times in March 2026, capping a month that also included 14-plus product launches and an accidental Claude Mythos leak. The incidents occurred on March 2, March 11, March 20, March 27, and March 28, with the final outage lasting nearly five hours — the longest recorded downtime in Claude’s operational history.
The March 2 incident hit on a Monday morning, with thousands of users reporting problems accessing Claude services. The outage affected Claude.ai and Claude Code, though Anthropic stated the Claude API was working as intended. Most users experienced authentication failures when attempting to log in, suggesting the issue was in the identity and session management layer rather than the model serving infrastructure.
On March 11, Down Detector reports spiked to over 1,400 at 10:29 AM ET. Anthropic confirmed the issue approximately two hours later and marked it as resolved shortly after. The March 20 disruption was different in character — starting at approximately 20:00 UTC, Claude.ai responses appeared to “hang” for around five seconds after text finished streaming before the message completed, a latency issue rather than a full outage.
Why It Matters
Five outages in a single month matters because Claude is no longer just a consumer chatbot. With $19 billion in annualized revenue and enterprise API market share that has grown from 12% to 32%, downtime now affects production workflows across thousands of companies. Claude Code alone generates $2.5 billion annually, meaning each hour of downtime carries measurable economic cost for development teams that depend on it for daily coding work.
The timing amplified the impact. Anthropic’s aggressive March launch schedule — more than 14 products and features — put additional load on infrastructure that was already absorbing a wave of new users. The combination of rapid product expansion and surging demand exposed infrastructure capacity limits that had not been visible at lower traffic levels.
Technical Details
The services affected across the five incidents included Claude.ai (web, desktop, and mobile applications), Claude Code, Claude Cowork, the Developer Console, and MCP calls. The core Claude API remained functional during at least some of the incidents, suggesting the disruptions were concentrated in the consumer-facing and developer tool layers rather than the underlying model serving infrastructure.
On March 27, users reported elevated error rates particularly affecting the Opus 4.6 and Sonnet 4.6 models, indicating model-specific routing or capacity issues rather than a blanket infrastructure failure. Anthropic doubled off-peak usage limits across Free, Pro, Max, and Team plans from March 13 through March 28, indicating the company was aware of capacity constraints and actively attempting to redistribute load across time zones.
Anthropic acknowledged the outages occurred but stated that “a detailed root cause analysis is still pending.” No post-mortem has been published for any of the five incidents as of the end of March.
Who’s Affected
The outages hit three distinct user groups with varying severity. Individual Claude.ai users faced login failures and response delays, disrupting personal and professional workflows. Claude Code users experienced interrupted coding sessions and lost context in long-running agentic tasks, which is particularly costly given the tool’s role in multi-step development workflows. Enterprise API customers were partially shielded during some incidents but reported elevated error rates on March 27.
A clear pattern of user migration from ChatGPT toward Claude was observed beginning in early March, following OpenAI’s Pentagon deal announcement. The influx of new users aligns with the timing of the first outage on March 2. Whether this surge directly caused the infrastructure overload has not been officially confirmed by Anthropic, but the correlation is difficult to dismiss given the company’s own acknowledgment of capacity constraints during this period.
What’s Next
Anthropic has not published a detailed post-mortem for any of the five March incidents. The company’s pending root cause analysis will be closely watched by enterprise customers evaluating Claude’s reliability for mission-critical workloads. The March outages stand as a concrete data point about the gap between Anthropic’s commercial growth trajectory and its infrastructure capacity to support that growth at the current scale of demand. Until Anthropic demonstrates that its infrastructure can handle both its existing user base and continued rapid growth without repeated disruptions, the reliability question will remain open.
