On the morning of Monday, March 2, 2026, thousands of users found themselves locked out of Claude. Anthropic's status page confirmed a widespread disruption tied to the login and logout paths on Claude.ai and Claude Code, while the underlying Claude API kept running. The company said it had identified the problem and was rolling out a fix, but never published a root cause. The timing made it sting: the outage landed just days after Claude climbed to the top of the App Store, in the wake of Anthropic's contentious negotiations with the Pentagon. A product surging in popularity got its first hard test of infrastructure reliability under a wall of new traffic — and briefly buckled.
For most people this was an inconvenience. For anyone whose brand visibility now flows through AI assistants, it was a preview of a structural risk worth taking seriously.
Key takeaways
- Claude.ai and Claude Code went down Monday morning on a login/logout fault; the Claude API stayed up, so programmatic access survived while human sign-in did not. - The outage followed Claude's rise to the App Store's number one spot, meaning the failure hit at peak load and peak attention. - When one assistant becomes a primary way people discover and buy, an outage doesn't just frustrate users — it redistributes their questions to other engines. - For brands, this is a concentration-risk lesson: visibility staked on a single AI surface is fragile, and share of voice should be measured across ChatGPT, Gemini, Perplexity, Copilot, Grok, and Google AI, not one of them. - Resilience in the agentic era is a portfolio problem — you want to show up wherever the query lands next.
What actually broke
The failure was narrow in shape but wide in effect. Login and logout are the front door; when that door jams, it doesn't matter that the rooms behind it are fine. Claude.ai users couldn't get in, and Claude Code — the developer-facing tool many teams now lean on daily — was caught in the same fault. The Claude API stayed healthy, which tells you the models and inference layer weren't the problem. This was an access and authentication issue, the kind that scales badly precisely when a service is most popular.
That last point matters. Anthropic hadn't just gained users; it had gained them fast, on the back of a public standoff over how its models could be used and a chart-topping app. Rapid, concentrated growth is exactly the condition under which authentication systems get stress-tested in public.
Outages move the map
Here's the part brands tend to miss. When a dominant assistant goes dark, the demand behind it doesn't evaporate — it reroutes. A shopper who was about to ask Claude for a product recommendation asks ChatGPT or Gemini instead. A developer who wanted Claude Code opens Cursor or Codex. Every outage is a small, involuntary experiment in which users get pushed onto the next engine, and whichever brands are well-represented there capture attention they didn't earn that day.
