Your Anthropic API key leaked.
Here's the fix, in order.

Step 1: Revoke and replace the key

One thing that makes this easy: Anthropic key rotation is two moves. You create a new key, and you delete the old one. There is no single "rotate" button, and you don't need one.

Go to the API keys page in the Claude Console: platform.claude.com/settings/keys. (Older console.anthropic.com links redirect to the same place.) Find the key that matches the leaked prefix.

If your app is live in production, create the replacement first so you can swap with almost no downtime:

1. Create a new key and copy it immediately. It is shown once, at creation, and never again.
2. Put it everywhere the old one lived: your local .env as ANTHROPIC_API_KEY, your CI secrets, and your host's environment variables (Vercel, Netlify, wherever you deploy). Redeploy.
3. Now delete the leaked key in the console.

If nothing is running in production, flip the order: delete the leaked key first, then create the new one at your own pace.

Finally, confirm the old key is actually dead. From a terminal:

curl https://api.anthropic.com/v1/models \
  -H "x-api-key: PASTE_THE_OLD_KEY_HERE" \
  -H "anthropic-version: 2023-06-01"

You want a 401 back, with an authentication_error in the body. 401 is the "this key doesn't work anymore" error. It's the sound of the door locking.

Step 2: Check usage for abuse

Open the usage page in the Claude Console (platform.claude.com/usage) and read the last 7 days. You're looking for anything that isn't you: spikes at hours you weren't working, models you never call, token volume that doesn't match your app's traffic.

Be honest with yourself about the timeline. If the key sat in a public repo for more than a day, assume someone found it. Automated scrapers hunt fresh public commits for key patterns around the clock. "Nobody saw my repo" is not a plan.

If you find spend you don't recognize:

• Revoke first (step 1), investigate second. Never the other way around.
• Contact Anthropic support with the dates, the leak timeline, and what you found. There's no guaranteed refund for leaked-key abuse, but a clear, honest account of what happened costs you nothing and gives them something to work with.
• Check your limits page (platform.claude.com/settings/limits). The console shows your organization's spend and rate limits there, so you know the worst-case number a stolen key could have reached.

One habit that shrinks the next blast radius: Anthropic's console supports workspaces, which let you segment API keys and control spend per use case. A separate key per environment, each in its own workspace, means a leaked dev key can't quietly drain your production budget.

Step 3: Rotate the neighbors

Here's the uncomfortable part. An sk-ant- key almost never leaks alone. It leaks because a .env file got committed, and that file usually holds the whole family: a database URL, a Stripe key, an OpenAI key, a Supabase service role key, session secrets.

If the Anthropic key was exposed, everything next to it was exposed. Each of those is its own small incident, and each gets rotated at its own provider's console. Work in this order:

1. Money and data first. Payment keys and database credentials. These are the ones that turn a bad morning into a bad quarter.
2. Write access second. Tokens that can push code or change infrastructure, like GitHub tokens.
3. Everything else third. Other AI provider keys, email tokens, analytics keys.

The moves are the same everywhere: open the API keys page in the provider's console, create a replacement, update your environments, kill the old credential, then confirm the old one fails. We keep a general runbook for this at what to do when any API key leaks, with the same calm-ordered approach as this page.

Step 4: Scrub git history. Last, and here's why.

A lot of people do this step first, feel better, and stop. That's the trap this page exists to prevent.

Deleting the file in a new commit does nothing. The old commit still contains the key, and anyone can read old commits.

Rewriting history helps less than it feels like it does. Tools like git filter-repo or the BFG can rewrite your repo so the key never appears, and a force-push replaces what's on GitHub. But git is a copying machine. Forks, clones, and anything a scraper already downloaded keep the old history. You cannot un-publish a secret. That's why the key dies at the provider in step 1, and the repo cleanup is hygiene, not the fix. Scrubbing alone is theater.

It's still worth doing, for two honest reasons: future clones of your repo stay clean, and nobody re-leaks the key by referencing an old commit. While you're in there, check whether any env file ever touched history:

git log --all -p -- .env .env.local .env.production | head

If any .env file shows up in history at all, every key that was ever inside it needs the step 1 treatment. Git history doesn't forget.

Then make the same mistake impossible to repeat:

echo ".env" >> .gitignore
echo ".env.local" >> .gitignore
echo ".env.production" >> .gitignore
echo "!.env.example" >> .gitignore

Step 5: What gets caught automatically (and what doesn't)

There is a real safety net here, and it's worth understanding exactly where its edges are.

Think of the automatic net as a smoke alarm, not a sprinkler system. It's genuinely good that it exists. It also goes off after the smoke.

FAQ

What happens if my Anthropic API key leaks?

Anyone who finds it can call the Claude API as you, which means spending your credits and running up your bill. If the key was pushed to a public GitHub repo, GitHub's secret scanning forwards exposed Anthropic keys to Anthropic, which revokes compromised keys and notifies you. Don't wait for that. Revoke it yourself, check usage for activity that isn't yours, and rotate whatever leaked alongside it. Keys leak in packs.

Does Anthropic automatically revoke leaked keys?

Sometimes, through one channel. Anthropic is a GitHub secret scanning partner, so keys found in public GitHub repositories get forwarded to Anthropic, which revokes them and notifies the affected user. That covers public GitHub only. A key leaked in an unscanned private repo, a frontend bundle, a screenshot, or a chat message has no one coming for it, and no auto-revocation undoes spend that already happened. Treat it as a backstop, not a plan.

How do I rotate an sk-ant key?

Open the API keys page in the Claude Console. Create the new key first and copy it right away (it's shown once). Update every environment: local .env, CI secrets, deployed host. Delete the leaked key. Then call the API with the old key and confirm you get a 401 authentication error back. About 10 minutes, start to finish.