AIBRIEF
The secure login tool inside most servers and clouds patches a flaw in its key holding helper, ssh agent, that could let a remote machine use login keys meant to stay local — and says it will release fixes faster as AI assisted bug reports surge.
OpenSSH, the secure-login tool embedded in most servers, clouds, and developer machines, is moving to more frequent, smaller releases after observing that AI-generated vulnerability reports are being independently rediscovered by other researchers. The team reads that pattern as evidence adversaries are likely finding the same bugs without reporting them, which makes disclosure-to-patch time itself a security variable. The shift lands with OpenSSH 10.5, released Tuesday.
The release ships three security fixes, per the txt advisory. The most consequential sits in ssh-agent(1), the helper that holds login keys: through the session-bind@openssh.com extension, a locked or restricted agent could be coaxed into letting a remote caller add hardware tokens or use forwarded keys meant to stay local. Brian Mingus of Cognatory reported and fixed a separate ssh(1) client use-after-free during a remote forwarding race, and Erichen of the Institute of Computing Technology, Chinese Academy of Sciences, reported that sshd's authorized_keys "restrict" keyword was not being applied to tunnel forwarding. No CVE numbers are listed in the release notes.
Portable builds now require ECC support in libcrypto, including NISTP521, satisfied by default in OpenSSL, LibreSSL, BoringSSL, and AWS-LC. --without-openssl builds are unaffected. Security reports go to openssh@openssh.com.