Browserjack can control authenticated browser sessions through OpenAI's locally installed browser runtime. Treat every report that touches session integrity, credential exposure, sandbox scope, or signature verification as security-relevant.

Use GitHub private vulnerability reporting for this repository. Do not open a public issue for exploitable problems.

You can expect an initial response within 7 days. Please include the Browserjack version, the ChatGPT.app version, and redacted browserjack doctor --json output.

In scope:

• Bypass of the fail-closed compatibility manifest
• Bypass of code-signature or Team ID verification
• Escape from the documented sandbox write scope
• Leakage of browser content, cookies, tokens, or form data through Browserjack's process chain, stdout, or stderr
• Tampering with the installed release, shim, or MCP registration

Out of scope:

• Vulnerabilities in OpenAI's ChatGPT.app, extension, or native host (report to OpenAI)
• Vulnerabilities in Claude Code (report to Anthropic)
• Attacks requiring an already-compromised local user account

Only the latest published release receives security fixes. The compatibility manifest intentionally rejects unknown ChatGPT.app builds; do not patch around it.