Flagsmith vs mod_auth_openidc

The Apache OIDC module is in the middle of the most consequential ten days in this window: 2.4.20 on August 1 rebuilt session and cookie encryption around PBKDF2-HMAC-SHA256 key stretching and invalidated every existing session, 2.4.20.1 on August 9 disclosed eight security issues found in an internal audit, and 2.4.20.2 the next morning walked back part of the fallout. The audit findings are not peripheral — the most serious lets a client inject headers that a protected backend trusts as authenticated identity claims on paths the module answers without authenticating. The 2.4.19.x line before it had already shipped an out-of-bounds read and write in the state-cookie parser.

The project has shifted from feature work to hardening its own attack surface, and it is doing so on its own initiative rather than in response to external reports. That posture has a cost operators are absorbing directly: two backwards-incompatible session format changes in six months, both of which log every user out on upgrade. The 2.4.20.2 release also shows the hardening overshooting and being corrected — unconditional secret masking made debugging impossible, so an opt-in escape hatch was added with a startup warning attached, and a derived-object cache tier added only weeks earlier was removed outright.

Expect the 2.4.20.x line to keep absorbing follow-up fixes from the same audit, and any further hardening to arrive with an explicit opt-out after the masking reversal showed operators cannot troubleshoot a protocol exchange they cannot read.