Authlib's JWS verification does not enforce the RFC 7515 'must-understand' semantics for the `crit` header parameter. Specifically, when a token includes unknown critical headers such as `bork` or security-sensitive headers like `cnf`, Authlib 1.6.3 verifies the signature and returns the payload without rejecting the token or enforcing the semantics of these critical headers. This behavior contrasts with strict verifiers like Java Nimbus JOSE+JWT and Node jose, which reject tokens with unknown critical headers. In heterogeneous deployments, this discrepancy can cause a gateway to reject a token while a backend service using Authlib accepts it, leading to potential authorization bypass, replay attacks, or privilege escalation. The vulnerability is identified as CVE-2025-59420 with a CVSS 3.1 score of 7.5 (high severity).

The vulnerability violates the JWS specification by accepting tokens with unknown critical headers, which should be rejected. This non-compliance can cause authorization policy bypass in environments where critical headers convey mandatory security semantics such as token binding (`cnf`). Services relying on Authlib 1.6.3 for JWS verification may accept malicious tokens rejected by other strict verifiers, resulting in split-brain acceptance. This can lead to replay attacks or privilege escalation if critical header semantics are ignored.

A patch is available for this vulnerability. Users should upgrade Authlib to a fixed version that correctly enforces the RFC 7515 'must-understand' semantics for critical headers. Until patched, avoid relying on Authlib 1.6.3 for JWS verification in security-sensitive contexts involving critical headers. No vendor advisory content contradicts this guidance.