In short: we tell vendors first, give them time to fix, and publish once users can protect themselves. Our default disclosure window is 90 days.
What this policy covers
- Vulnerabilities the AI Threat Lab finds in AI models, applications, agents and related products.
- Vulnerabilities in AI systems reported to us by outside researchers who want help coordinating disclosure.
- Vulnerabilities in the initiative's own websites, learning platform and lab infrastructure.
When we find a vulnerability
Confirm
We reproduce the issue and assess its severity and who is affected.
Notify
We report it privately to the vendor or maintainer, with the details needed to fix it.
Coordinate
We work with the vendor on a fix and a publication date. Our default window is 90 days from notification.
Publish
We publish an advisory with guidance for defenders, crediting the vendor's response.
We may shorten the window if a vulnerability is being actively exploited, or extend it when a vendor is making good progress on a complex fix. If a vendor does not respond, we may publish after the window with guidance that helps users protect themselves.
How to report a vulnerability
Use our contact form and choose "Vulnerability disclosure". Give a short summary only. Do not include exploit code or sensitive data in the first message; we will arrange a secure channel for the details.
A useful report includes:
- The affected product, version or service.
- A description of the issue and its impact.
- Steps to reproduce, which we will request through the secure channel.
- Whether you want credit, and the name to use.
Our commitments to reporters
- Acknowledge your report within five business days.
- Keep you informed as we validate and coordinate.
- Credit you in the advisory, if you wish.
- Not share your identity with a vendor without your permission.
Good-faith research on IIAS systems
If you research the initiative's own systems in good faith and in line with this policy, we will not pursue legal action against you, and we will work with you to understand and fix the issue. Good faith means you:
- Access only the data needed to show the issue, and do not keep or share it.
- Do not degrade our services or affect other users.
- Give us reasonable time to fix the issue before you disclose it.
Out of scope
- Denial-of-service testing.
- Social engineering of staff, learners or members.
- Physical attacks.
- Lab environments used for their intended exercises. Lab vulnerabilities that let you affect other learners or IIAS systems are in scope.