AI Vulnerability Database

Source-linked intelligence for AI frameworks, runtimes, agent platforms, model tooling, and supporting software.

Find a product or identifier.

Search by CVE, AASD identifier, product, vendor, component, or security term. Results open in the full database.

Evidence model

Every record identifies whether Advent maintainers reviewed the evidence or whether the entry came from the automated upstream feed.

Maintainer reviewed

These records include reviewed scope, affected versions, attack conditions, impact, mitigation guidance, and primary public evidence.

Automated intelligence

These records reproduce normalized information from authoritative upstream sources. They are not independent exploitability decisions.

Safe publication

The database excludes zero-days, stolen data, live-target instructions, and turnkey weaponized payloads.

Apply the intelligence

Advent can review the affected architecture, exposure, controls, and launch evidence as part of an AI security review.