Offensive AI · 7 findings
AI-Found Vulnerability Ledger
Vulnerabilities whose discovery a public source credits to an AI system: what AI found, in which software, how severe and which system found it. Every entry links to the crediting source and quotes it word for word.
Methodology · API: /api/v1/ai-found-vulns?system=&since=
Credited findingsall time
7
Credited in last 30 days30d
7
Critical or highseverity
4
AI systems creditedsystems
5
In CISA KEVexploited
0
Findings
7 shown| Date | Vulnerability | Product | Severity | AI system | Source & evidence |
|---|---|---|---|---|---|
| 2026-11-20 | CVE-2026-99758 Linux kernel SCTP sctp_init_cause() missing chunk padding check causes skb overflow | Linux kernel | Not rated | unnamed AI system | CVE-2026-99758: In the Linux kernel, the following vulnerability has been resolved: sctp: account for chunk padding in sctp_init_cause() sctp_init_cause()… |
| 2026-11-19 | CVE-2026-126714 Remote code execution via PHP default file-based session handler (session upload progress / large strings) | PHP | Critical 9.2 | unnamed AI system | How Cloudflare WAF helps protect against phpAfterShell and other critical PHP flaws Evidence“depthfirst says it discovered and reported phpAfterShell. The CVE record credits Leo Lin and Asahi as reporters.”ThreatDeck story |
| 2026-11-18 | CVE-2026-123007 Unauthenticated remote code execution in NGINX, chainable with a Linux kernel bug to reach root | NGINXF5 | Critical | Xint AI harness Xint | It’s NGINX’s turn to cause vulnpocalypse headaches Evidence“others have targeted NGINX with AI pentesting, but none found this critical bug, which is a testament to our team and the engineering expertise that's gone into our AI harnesses.”ThreatDeck story |
| 2026-11-18 | no id yet Unauthenticated remote code execution in NGINX | NGINXF5 | Not rated | Xint AI harnesses Xint | It’s NGINX’s turn to cause vulnpocalypse headaches Evidence“Others have targeted NGINX with AI pentesting, but none found this critical bug, which is a testament to our team and the engineering expertise that's gone into our AI harnesses.”ThreatDeck story |
| 2026-11-17 | CVE-2026-114938 Memory corruption flaw in Java allowing pre-authentication takeover of servers (Java4Shell) | JavaOpenJDK | Critical | Codex OpenAI | CISA warns of ‘Java4Shell,’ an AI-found Java bug touching most of the Fortune 500A vulnpocalypse bug worthy of attention: Researchers disclose critical flaw in Java with help of AI |
| 2026-11-16 | no id yet Previously unknown vulnerability used to access Hugging Face's servers during OpenAI agent sandbox escape | Hugging Face servers | Not rated | unnamed AI system OpenAI | California Sues OpenAI, Seeking Court-Ordered Changes After the Hugging Face Hack Evidence“OpenAI said its AI used stolen credentials and discovered a previously unknown vulnerability to access Hugging Face’s servers.”ThreatDeck story |
| 2026-11-05 | 81 flaws High-severity Firefox vulnerabilities discovered by GPT-6 Cyber during security testing | FirefoxMozilla | High | GPT-6 Cyber OpenAI | OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugsOpenAI debuts GPT-6 Cyber, touting better performance — and safety guardrails Evidence“GPT-6 Cyber discovered 81 previously unknown high-severity vulnerabilities in Firefox” |
Methodology
Discovery credit is as stated by the cited source; ThreatDeck does not independently verify it. Entries need an explicit statement that an AI system found the vulnerability, quoted verbatim.
- Credited by the cited source; not independently verified. We record what the source says, not our own assessment.
- An entry needs an explicit statement that an AI system, agent or model (or AI-assisted research) found the vulnerability. Attackers using AI, flaws in AI products found by people, and generic claims are left out.
- The evidence quote is checked word for word against the stored source text. When the source names no system, the entry says “unnamed AI system”.
- Batch disclosures without individual ids are one entry with a count of confirmed vulnerabilities. CVSS, severity and KEV come from NVD, GitHub advisories and CISA KEV when a CVE is known.
- Date is the stated discovery date when the source gives one, otherwise the source's publication date.