Mythos 'Discovered' a CVE in Its Training Data and That's Still Worrying
Summary
The Rival article analyzes how Claude Mythos reportedly discovered CVE-2026-4747 in FreeBSD's RPCSEC_GSS and discusses the broader implications for AI-assisted vulnerability discovery, training-data leakage, and patching. It situates the finding in a historical context around ONC RPC/NFS and Kerberos, highlighting how legacy flaws persist and how AI might accelerate exploitation unless defenses are tightened.