Removed: roughly 4,500 lines of post-quantum key code from sshd’s pre-authentication path — about 16% of its pre-auth code in a default build — by moving the allowlist check ahead of the parse. No vulnerability in that code was patched.
Outcome: OpenSSH 10.5 shipped the reorder and credits “Christopher Paul Rohlf of Anthropic”. Anthropic’s three passes reported 493 issues; in the eleven projects it gated, 93 pre-authentication surfaces did work nothing asked for, 37 ran code or wrote state before authentication, and 61 configuration combinations quietly weakened a pre-auth control.
Anthropic ran three passes over 25 open-source projects — more than five million lines of code and over 17,000 configuration options: what is reachable in the default configuration that no feature needs, what executes before authentication, and which configuration combinations silently switch a control off. It reported 493 issues. In the eleven gated projects, 93 pre-authentication surfaces did work nothing asked for, 37 ran code or wrote state before authentication, and 61 configuration combinations quietly weakened a pre-auth control; across the set, 94 of 162 undocumented option combinations turned a security control off with no error and no warning (Anthropic, Using Claude to reduce the attack surface of code).
The OpenSSH finding is worth reading in full. Roughly 4,500 lines of post-quantum key code — about 16% of sshd’s own pre-authentication code in a default build — were reachable by an unauthenticated client even when the administrator had excluded the algorithm, because the key type was parsed before it was checked against the allowlist. OpenSSH 10.5 moved the check ahead of the parse and credits “Christopher Paul Rohlf of Anthropic”, noting that the change “removes at least some key parsing and verification paths from the pre-auth attack surface.” No vulnerability in that code was patched. The code stopped being reachable, which was cheaper and covers everything nobody has found yet. Anthropic’s own summary is this manifesto in one sentence: “a defender who reduces unnecessary attack surface in their code reduces the risk from all of the unknown vulnerabilities in that code at once—including any that are already being quietly exploited.”
Sources:
- Anthropic, Using Claude to reduce the attack surface of code (20 October 2026)
- OpenSSH, Release 10.5 (11 August 2026)