The vulnerability is an Algorithm Confusion / Key Confusion attack.
The server verifies JWTs like this:
jwt.verify(token, KEY)
But it never restricts allowed algorithms.
That means an attacker can:
- Change alg from RS256 β HS256
- Use the PUBLIC key as the HMAC secret
- Forge arbitrary tokens like:
{ "role": "admin" }
Impact:
- Full auth bypass
- Privilege escalation
- Account takeover
- Admin access without private key compromise
Fix:
jwt.verify(token, KEY, { algorithms: ["RS256"] })
Saw something interesting today about #Anthropic and #ClaudeCode.
There are claims of a potential data leak β not sure how accurate this is yet.
If true, this raises serious questions about AI system security.
Anyone else following this?
https://t.co/fnXUwHHgDj