ANI (Adversarial Neural Inspector) is itself a security tool. If you find a security issue in ANI itself — not in a target application you scanned with it — please follow this policy.

Do not open a public GitHub issue for security bugs. Public disclosure gives attackers a head start on a fix.

Report privately to:

• Email: abhirupguhakolkata@gmail.com
• GitHub: use GitHub's private vulnerability reporting on the repository's Security tab.

Please include:

1. A short title and the affected component (CLI, sidebar, reporting, etc.).
2. Steps to reproduce, or a proof-of-concept if the issue requires a real target.
3. The impact you observed (XSS, credential leak, RCE, etc.).
4. The version of ANI and the Python / Node / Firefox versions you tested on.
5. Whether the issue requires authentication or specific permissions to exploit.

You can encrypt sensitive details with this maintainer's PGP key if needed; if not, plain text is fine for an initial report and we can move to a signed channel from there.

• Acknowledgement: within 7 days of the report.
• Triage: within 14 days, with a severity assessment (CVSS or qualitative) and a fix plan.
• Fix timing: critical issues target a patch release within 30 days; high within 90 days; lower-severity issues are bundled into the next regular release.
• Credit: credited in the release notes unless you ask to remain anonymous.
• Coordinated disclosure: a 90-day disclosure deadline is the default. We will agree on an extension if a fix needs more time.

Only the latest release on master receives security patches. Older versions are not patched; please upgrade.

These are issues in the targets ANI scans, not in ANI itself, and should not be reported here:

• Vulnerabilities found in third-party AI chat applications you scanned with ANI.
• Issues in third-party dependencies (Playwright, Typer, Rich, etc.) — report upstream.
• Reports of ANI "succeeding" against a chat app you own — that is ANI working as designed; remediate your target.

ANI ships with several built-in protections. Do not disable them when reporting issues — flag the case where they don't apply.

• HTML reports use Jinja2 autoescape and a Content-Security-Policy meta tag.
• Sidebar results are built with createElement + textContent (no innerHTML).
• Credentials at rest are Fernet-encrypted (auth_profiles/, sessions/).
• Network-captured headers are redacted before any serialized output.
• The deprecated --cookie flag emits a warning at use; prefer --cookie-file or ANI_AUTH_COOKIE.

ANI is developed and reviewed with:

• A regression test suite of 46 tests including explicit XSS-regression, encryption, and PII-redaction tests.
• Manual code review for any new payload, browser, or auth surface.
• No telemetry, no auto-update, no remote code fetch at runtime.

If you spot a gap, please report it through the channels above.