Vulnerability  ·  2026-10-24

Goose terminal integration sends shell command history verbatim to LLM endpoints → unredacted credential disclosure (CVE-2026-111079)

VulnerabilityMedium impactGlobalCVE-2026-111079
A flaw in Goose: with terminal integration enabled, shell command history is transmitted verbatim to LLM provider endpoints without sanitization. If a user enters sensitive data (API tokens, passwords, env-variable secrets) in the terminal, it travels across trust boundaries and into provider logs. CVSS 7.4 (changed scope, confidential impact), published 2026-10-22 via Red Hat.
Local agent tools that surface raw shell history to an LLM provider turn the LLM query path into a credential disclosure channel. For AI deployments with agents driving terminal/CLI work, this leaks the same secrets the agent can see.
When the terminal integration feature is enabled, Goose captures shell command history and sends it verbatim without sanitization to the configured LLM provider during queries, leaking anything the user typed — including credentials embedded in commands or environment variables (CWE-497).
Goose (shipped in Red Hat Enterprise Linux 9/10; the Goose terminal-integration feature) — terminal history sent to LLM provider endpoints
Red Hat advisory CVE-2026-111079 / bug 2554025; update Goose to a fixed build, disable terminal integration where not needed, and avoid placing secrets in shell history. Review provider-side logs for leaked credential strings.
NVD entryRed Hat CVE advisoryRed Hat bugzilla
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →