CVE-2026-45792 describes an insufficient verification of data authenticity in the rtk tool before version 0.32.0. rtk automatically loads a project-local filter configuration file (.rtk/filters.toml) with the highest priority and without user notification. An attacker with the ability to place files in the working directory can supply a malicious filter that applies regex-based modifications to shell command outputs before they are presented to the language model context. This can selectively suppress or alter command output, potentially hiding malicious code during AI-assisted development or review. The vulnerability is addressed by fixing this behavior in rtk version 0.32.0.

An attacker can manipulate the command output seen by the language model by injecting malicious filter files in the project directory. This can lead to concealment of malicious code or other important information by selectively suppressing or altering command outputs such as file contents, diffs, and security scan results. This undermines the integrity of AI-assisted development or review processes. The CVSS 4.0 score is 6.9 (medium severity), reflecting the local attack vector with user interaction required and high impact on confidentiality and integrity.

Upgrade rtk to version 0.32.0 or later, where this vulnerability is fixed. Prior versions improperly trust local filter configuration files without user notification, enabling this attack. No other official remediation or temporary fixes are documented. Users should avoid running vulnerable versions in untrusted directories until patched.