Field note
Claude Not Working: Outage or Capacity?
When Claude stops responding, the message on screen tells you whether it is an outage, a capacity squeeze, your own usage limit or a problem on your side.
When Claude stops responding, the message on screen tells you whether it is an outage, a capacity squeeze, your own usage limit or a problem on your side.
Symptoms and what they usually mean
"Claude is not working" covers four different situations. Before anyone changes an account, a network rule or a plan, read the exact wording and match it to the table.
| What you see | Most likely | First action |
|---|---|---|
| Page will not load, or every request errors for everyone in the office | Incident | Check the status page |
"Opus is experiencing high load", 529 Overloaded, or a model-specific message that clears on retry | Capacity | Wait a few minutes, retry, or switch model |
| "You've hit your session limit" or "You've hit your weekly limit", with a reset time | Usage limit | Wait for the reset or use usage credits |
| "You've hit your individual spend limit" or "org's monthly spend limit" | Spend cap you set | Admin raises the cap or you wait |
unable to get local issuer certificate, login loops, 403 Forbidden, works on mobile data but not the office Wi-Fi | Local, network or identity | Your IT, not Anthropic |
| One person affected, everyone else fine | Account or device | Sign out and in, check the plan and authentication |
Check status before anything else
Anthropic publishes a status page at status.claude.com, which also answers at status.anthropic.com. At the time of writing it lists five components: claude.ai, Claude Console, Claude API, Claude Code and Claude Cowork. Each shows Operational, Degraded Performance, Partial Outage, Major Outage or Maintenance, and there is an incident history you can scroll back through.
Two habits save time:
- Subscribe. The page offers email, SMS, Slack, Microsoft Teams, webhooks and RSS. One person in IT subscribing to Slack alerts means the helpdesk knows about an incident before the first ticket arrives.
- Match the component. A Claude Code incident does not mean the chat app is down, and an API incident may not affect Team plan users at all. Tell staff which product is affected.
If status is green and the whole office is failing, the problem is probably between you and Anthropic: a proxy, TLS inspection, a DNS change or a conditional access policy. Test from a phone on mobile data. If that works, hand it to IT.
Capacity is not an outage
A capacity message means the model is busy, not broken. Claude Code shows API Error: Repeated 529 Overloaded errors or a model-specific high-load notice. These clear on their own, often within minutes.
What to do:
- Retry once or twice, spaced out. Do not hammer the button; that makes the queue longer for everyone.
- Switch model for the task in hand. In Claude Code use
/model; in the chat app pick a different model from the selector. A capacity notice for one model family does not block the others. - Note the time. If capacity messages recur at the same hour every day, schedule heavy work outside that window.
Do not respond to a capacity message by buying a bigger plan. It will not change the queue.
Usage limits look like failures but are not
A limit message names a reset time and is shared across Claude chat, Cowork and Claude Code on the same plan. Run /usage in Claude Code or open Settings, Usage on claude.ai to see the five-hour and weekly bars. The model-specific variant, such as "You've hit your Opus limit", does allow switching to another model family with /model.
If limits are a weekly event rather than a surprise, that is a planning problem. Our notes on usage credits and the five-hour reset cover the options.
What IT should check
- Proxy and TLS. Corporate inspection that re-signs certificates produces
unable to get local issuer certificatein Claude Code. Configure the corporate CA rather than bypassing verification. - Identity. SSO changes, a disabled account or an expired session produce login loops that look like an outage to the user.
- Environment variables. A developer with
ANTHROPIC_API_KEYset in their shell is billed to the API, not the subscription, and can hit a Console rate limit while colleagues are fine./statusin Claude Code shows which authentication is active. - Version.
claude --version, then/doctor, before raising anything. - A written continuity note: which tasks can wait, which have a manual fallback, and who decides. Our AI chatbot usage limits and business continuity note is a starting point.
When to escalate to Anthropic
Escalate when status is green, mobile data behaves the same as the office network, the message is not a limit, and /doctor is clean. Use Get help from the lower left on claude.ai. At the time of writing, Pro and Max subscribers, Team and Enterprise owners and Console admins can reach a human specialist; other Team seats need the owner to raise it. Include the work email of the affected person, the exact error, a screenshot with no customer data in it, and when it started. For billing questions on the Console, sign in at platform.claude.com and use the same Get help route.