Vulnerability / research
Goose terminal integration sends unredacted shell history to LLM providers
Reported
Published here
What happened
Red Hat researchers reported that Goose terminal integration forwards recent shell commands to the configured LLM endpoint without redaction, potentially disclosing API keys, passwords and secret-bearing environment assignments.
Parties and reporting
AI tools and systems: Goose terminal integration (goose term / @goose), Configured LLM provider endpoints
Outcome
CVE-2026-111079 identifies an information-disclosure flaw. The researcher report lists versions 1.16.0–1.54.0; reviewed sources identify no fixed release or malicious exploitation.
Operating lesson
Minimize shell history sent to models, redact secrets before transmission, and isolate credential-bearing terminal activity from AI context collection.
Evidence and limits
The report was created privately October 16. Its restricted groups and embargo marker were removed October 22, matching the CNA public date. This is a disclosed data-handling vulnerability, not a confirmed intrusion or proof of provider misuse.
Sources
- Red Hat CNA record with October 22 public-disclosure date
Primary source ·
- Red Hat researcher report and exposure scenarios
Primary source ·
- Red Hat report history showing October 22 embargo removal
Primary source ·