What happened
A vulnerability has been identified in Goose, affecting systems where the terminal integration feature is enabled. This flaw is associated with the application capturing shell command history and transmitting it verbatim to configured Large Language Model (LLM) provider endpoints during query operations. The transmission occurs without sanitisation, which means that any sensitive data entered into the terminal—such as API tokens, passwords, or environment variables containing secrets—could be inadvertently disclosed. This data is sent across trust boundaries, potentially leading to sensitive information exposure. The vulnerability is classified as high severity, primarily because of the risk of credential leaks through unredacted command histories. It does not involve direct code execution or system compromise, but the information disclosure could facilitate further attacks if sensitive credentials are exposed.
Who is affected
Organisations running Goose with the terminal integration feature enabled on Red Hat Enterprise Linux 10 or Red Hat Enterprise Linux 9 are affected by this vulnerability. Systems where the application is configured to send command history content to external LLM providers without sanitisation are vulnerable. Despite the high severity, the vulnerability requires user interaction and content delivery via remote mechanisms, and is not dependent on exposed network services.
Recommended actions
- Implement content filtering, sandboxing, or other attack-surface reduction techniques on systems that process untrusted content, particularly those that handle terminal inputs or display command history data.
- Restrict or monitor the application’s interactions with external LLM provider endpoints to prevent or identify inadvertent transmission of sensitive information.
- Increase user awareness regarding the risks of entering sensitive data into terminal sessions when terminal integration features are enabled.
