Claude Token Theft: What Claude Code Users Must Do Now

What is happening with Claude token theft?

Hackers are stealing authenticated Claude browser sessions and using them to generate unauthorized Claude Code OAuth tokens, draining paid subscriptions like Claude Max. This isn't a billing glitch—it's a security issue. For example, one consultant saw usage climb from 45% to 55% while idle, even after disabling tools and tasks. Anthropic had to suspend the account and revoke sessions.

How do attackers steal Claude tokens?

Attackers use infostealer malware to copy browser cookies, saved credentials, and active-session data from infected computers. They don't need your password or 2FA code—a stolen session proves they're already logged in. This malware can come from downloaded software, malicious links, or compromised websites. Once they have your session, they can mint OAuth tokens for Claude Code, allowing continued access and usage consumption.

What are the warning signs of Claude account compromise?

Watch for these red flags:

• Usage limits rising while you're away from your desk
• A freshly replenished allowance disappearing quickly
• Claude Code capacity consumed when no tasks are running
• Unfamiliar account emails, forced sign-outs, or changed payment details
• Unrecognized usage-credit purchases or charges
• Agent output or logs showing work you didn't initiate

If you normally use 15-25% of your allowance daily, a 20% jump overnight while your laptop is closed should be treated as an incident.

What should I do if my Claude account is compromised?

Follow this checklist immediately:

1. Stop using the affected machine for Claude and sensitive logins. Disconnect from networks if you suspect malware.
2. Record evidence: screenshots of usage, timestamps, billing, alerts, and logs.
3. Contact Anthropic support to report unexplained consumption and request session invalidation.
4. Revoke sessions and tokens—not just passwords. Ask Anthropic to invalidate all active sessions and server-side Claude Code tokens.
5. Scan and clean your device before signing back in. Remove any malware first.
6. Secure your email from a known-clean device: change password, sign out of other sessions, enable 2FA.
7. Rotate other credentials stored in the browser, like banking, cloud, GitHub, and client accounts.
8. Review payments and usage credits. Disable auto-reload if you have it.
9. Rebuild trust carefully: log in from a clean device, reauthorize only needed tools.

How can I better secure Claude Code and agents?

Treat Claude like a production SaaS identity, not just a chat app:

• Use a separate browser profile for AI development and agent tools.
• Keep your browser, OS, and endpoint protection updated.
• Minimize browser extensions; only keep trusted ones.
• Never store long-lived tokens in plaintext files, shell history, or shared drives.
• Separate personal experimentation from client work using distinct accounts and environments.
• Limit agent permissions to only what's necessary.
• Maintain an activity log for automations, noting when agents run and expected token usage.
• Set financial guardrails: use low usage-credit limits and avoid unlimited auto-reload.
• Treat unusual token usage as a security alert and have a response playbook.

FAQ

Can 2FA protect me from Claude token theft?

No, 2FA doesn't help if an attacker steals your active browser session. They can bypass login entirely. That's why session hygiene and endpoint security are critical.

Is Claude itself the source of the malware?

No, Anthropic has no reason to believe the malware came from Claude. The issue is that infected endpoints leak session data used across many services.

What should I do if I see unexplained usage but no other signs?

Treat it as a potential compromise. Contact Anthropic support, record evidence, and follow the response checklist. It's better to be safe.

How can I monitor my Claude usage effectively?

Establish a baseline of your normal usage. Check your usage dashboard regularly, set up billing alerts if available, and review activity logs for your agents. If you see anomalies, investigate immediately.

If you run AI agents for your business, securing them is as important as securing your bank account. Need help setting up secure AI workflows? Get a free audit to identify risks in your automation setup.

Ready to talk it through? Get a free audit