OpenSSH 10.3 Fixes Shell Injection and Multiple SSH/sshd Security Flaws

OpenSSH 10.3 has been released with five security fixes affecting both the ssh client and sshd, led by a shell-injection risk in ssh(1) where attacker-controlled usernames could trigger arbitrary shell command execution in configurations that use %-token expansion such as %u inside Match exec blocks. The release also fixes improper certificate principal matching in authorized_keys, incorrect acceptance behavior around restricted ECDSA algorithms, and a legacy scp(1) issue that could preserve setuid/setgid bits on downloaded files when run as root under specific conditions.

The update changes certificate handling so empty principals sections no longer act as wildcard matches in risky scenarios, adds command-line validation for ProxyJump/-J to reduce shell-injection exposure, and includes a portability fix for a PAM-related username confusion issue introduced in the OpenSSH 10.2p1 stable branch. OpenSSH 10.3 also drops support for peers that do not implement rekeying and adds operational improvements including IANA-assigned SSH agent forwarding codepoints, per-source penalties for invalid usernames, better multiplexing diagnostics, and fixes for PKCS#11 PIN entry, FIDO/WebAuthn certificate signatures, and an sshd crash condition.