Low Severity
Published: Oct 31, 2026
Tracking CVEs discovered by AI systems — Anthropic, OpenAI, Google Big Sleep, Microsoft MDASH, DepthFirst
<p align="center"><a href="https://www.securin.io/"><picture><source media="(prefers-color-scheme: dark)" srcset="https://www.securin.io/images/securin-logo.png"><img alt="Securin" src="https://www.securin.io/images/securin-logo-dark.png" width="200"></picture></a><br><a href="https://www.securin.io/">securin.io</a></p>
# 🤖 AI-Discovered CVE Tracker
An evidence-backed list of CVEs credited to AI vulnerability-discovery systems — Anthropic, Google DeepMind (Big Sleep), OpenAI, Microsoft (MDASH), and DepthFirst AI.
**Jump to:** [Patch first](#patch-first) · [AI systems](#ai-systems-credited) · [Full CVE table](#cve-table) · [Contributing](#contributing)
## At a glance
- **597 published CVEs** are credited to five AI systems (3 more IDs are reserved and not yet public).
- **356** have a public source saying an AI system found the bug — 166 by the AI on its own, 190 by researchers working with it.
- **178** are credited to staff of an AI lab, but the source does not say AI was used; **14** are listed by the lab with no method stated; **49** have no public attribution evidence found.
- **5** are in CISA's Known Exploited Vulnerabilities catalog and **10** in Securin KEV (5 of them not yet in CISA KEV) — see [Patch first](#patch-first).
- By official CVE-record score: **70 Critical**, **269 High**, 206 Medium, 40 Low, 12 not yet scored.
*Last updated: 2026-10-27*
> **Disclosure:** this tracker is maintained by Securin. CVE IDs link to Securin's public vulnerability database. Attribution evidence, dates, CVSS scores and CWEs come from the official CVE records, vendor advisories and the AI labs' own disclosures — linked per row — so they can be verified independently.
## Patch first
CVEs in this list with evidence of exploitation in the wild. **CISA KEV** is the U.S. government's public catalog of confirmed exploited vulnerabilities; **Securin KEV** is Securin's own catalog, which also counts exploitation reported in threat intelligence and is not published as a pu