The latest tagged release is supported. Earlier versions receive only security fixes that backport cleanly.

Please do not open a public GitHub issue for security vulnerabilities.

Instead, email the maintainers with:

• a description of the issue
• minimum reproducer
• impact assessment
• your contact for follow-up

You will receive an acknowledgement within 72 hours and a planned-fix timeline within 7 days. If the issue is high-severity (RCE, auth bypass, data leak), the maintainers will coordinate a private fix and a coordinated public disclosure.

The gateway routes prompts that may contain proprietary data to multiple LLM providers. The most sensitive failure modes are:

1. Prompt-injection bypass — the pre-classifier or post-validator misses injection attempts and leaks system prompts or tool definitions.
2. Audit-log corruption — calls succeed but compression / cost metadata is wrong, so cost tracking lies.
3. Bridge auth leakage — a subscription bridge somehow exposes its OAuth token or session cookie via the audit log or response.
4. Dashboard auth bypass — endpoints under /api/dashboard/* serve real user data; the DASHBOARD_AUTH_TOKEN check must not be bypassable.
5. Build-drift exploit — running stale compiled code could re-introduce a previously-fixed vulnerability. The launch-time check-build-drift.mjs guard addresses this; if you find a way around it, please report it.

• Issues affecting only out-of-band external services (Anthropic API, OpenAI API) that the gateway proxies to but doesn't control.
• Social-engineering attacks on subscription accounts.
• Self-imposed loss of subscription quotas by misrouting.