kitty CVE-2026-95832

Requires kitty version 0.47.3 up to (not including) 0.49.0. The attacker must get crafted bytes written to the terminal that the user renders - e.g. displaying a malicious file, command output, log, SSH banner, or web response containing the color-control OSC query whose field name encodes a shell command using only printable ASCII (excluding ';' and '=' which are consumed as delimiters). A newline to submit the injected command is obtainable via the ssh kitten path (handle_remote_ssh/get_ssh_data emits leading newline bytes before credential checks). Limiting factors: local attack vector (AV:L) and required user interaction (UI:P) - the victim must actually display attacker-controlled output in an affected kitty build; command runs only with the user's own shell privileges, and default configurations are affected once malicious output is rendered.

This is a genuine high-priority terminal-emulator RCE, but the CVSS 4.0 score of 9.3 slightly overstates real-world reach because the input CVSS vector itself is AV:L/UI:P - exploitation is local and requires user interaction, not a remote unauthenticated one-click. The mechanism is well documented and corroborated by the vendor fix commits (03fbdad and 57bb144) and release v0.49.0: kitty's color_control() echoed an unrecognised query field name verbatim into an OSC reply, and write_escape_code_to_child() wrote that reply back into the pseudoterminal master where the shell reads it as if typed by the user, enabling command injection (CWE-74). Notably this is a regression/bypass of the CVE-2026-54057 fix: the 0.47.3 mitigation reduced the echoed payload to printable ASCII and consumes ';' and '=' as delimiters, but every other printable character survives, which is sufficient to compose a shell command, and a newline needed to submit the command is sourced from handle_remote_ssh()/get_ssh_data() before credential checks. No EPSS score, no CISA KEV listing, and no confirmed public exploit are present in the provided data, so this is not evidence of active exploitation; there is a CNA/third-party writeup (secur0.com) that may contain technical detail but POC status is unconfirmed. Net: treat as a real priority for kitty users (patch to 0.49.0) but not an emergency mass-exploitation event on current evidence.

A user running kitty (0.47.3-0.48.x) views attacker-controlled output - e.g. cats a malicious file, curls a hostile URL, reads a crafted log, or connects where attacker bytes reach the terminal - that contains a crafted color-control OSC query with a field name encoding a shell command. kitty echoes the field name back into the pty master, and because that reply is indistinguishable from typed input (with a newline obtainable via the ssh kitten path), the user's shell executes the attacker's command at the user's privilege level. No public exploit code is identified in the provided data, though a CNA writeup exists.