OpenID Connect Core through 1.0 errata set 2 allows audience injection in certain situations. When the private_key_jwt authentication mechanism is used, a malicious Authorization Server could trick a Client into writing attacker-controlled values into the audience, including token endpoints or issuer identifiers of other Authorization Servers. The malicious Authorization Server could then use these private key JWTs to impersonate the Client.
The authentication algorithm is sound, but the implemented mechanism can be bypassed as the result of a separate weakness that is primary to the authentication error.