Skip to content

Topic

AI-Driven Vulnerability Research

The use of AI models and automated tools to find, analyze, and exploit software vulnerabilities, reshaping security research and disclosure practices.

Current stories

securityOne report1 publisher

Vghost hands root to any local user on four of the five Linux kernels Tencent tested

Tencent Zhuque Lab published working demos on October 12 for Vghost (CVE-2026-74580), a vhost bug that gives local users root on Linux 4.11 through 7.2-rc4. The August 3 fix is out, but Ubuntu 24.04 and 26.04 kernels were still listed as in progress on October 13.

Reality

Evidence70
Adoption
Not scored
Hype gap−10
Incentives
Not scored
Confidence65
securityOne report1 publisher

Chainguard's Mythos harness finds a vLLM flaw that lets one client dictate another's output

Chainguard says its Mythos harness found a vLLM flaw, left in the public tracker for four months, that lets one client force another client's output. Versions before v0.31.0 are affected, and nothing is logged when the forced answer returns.

Publishers:chainguard.dev

Reality

Evidence58
Adoption
Not scored
Hype gap−5
Incentives65
Confidence60
securityOne report1 publisher

Microsoft's AI-driven FORGE lab helped find 140 Windows CVEs in five months

Microsoft's FORGE lab helped discover Windows flaws assigned 140 CVEs from May to September 2026, 52 of them fixed in the September release. Microsoft says the pace now depends on how fast its response center can validate and fix what the AI agents report.

Publishers:microsoft.com

Reality

Evidence40
Adoption30
Hype gap+15
Incentives70
Confidence40
buildOne report1 publisher

Cantina's open-weights exploit model ran a 60-task security eval for $2.38

Cantina released apex-flash-1, an open-weights vulnerability-research model it says solved 40 of 60 tasks for $2.38, against $74.68 for Claude Opus 5 High. There is no hosted endpoint, so teams download the 321-billion-parameter weights, pay for their own inference and verify the numbers themselves.

Publishers:runtimewire.com

Reality

Evidence45
Adoption
Not scored
Hype gap+35
Incentives70
Confidence40
buildOne report1 publisher

Cantina publishes a security model built from 50 of its own vulnerability cases

Cantina released Apex Flash-1, an open-weights security model fine-tuned on 50 of its own vulnerability cases. Its only performance evidence is a 60-task evaluation the company ran itself, and a separate build is modified to refuse fewer requests.

Publishers:runtimewire.com

Reality

Evidence40
Adoption
Not scored
Hype gap+15
Incentives70
Confidence40
securityConfirmed3 publishers

CISA reframes the CVE program around data quality as 2026 heads for 96,000 records

Disclosure volume is climbing faster than the process around it. CISA's answer is a framework that describes what a good CVE record is and how the program should be judged on producing one.

Perspective Coverage

3 publishers
Builder
Builder 33%
Operator
Operator 50%
Investor
Investor 17%

Reality

Evidence68
Adoption
Not scored
Hype gap+30
Incentives55
Confidence66
buildOne report1 publisher

Mythos Preview chains Cloudflare's invisible backlog bugs into one severe exploit

Cloudflare pointed Anthropic's Mythos Preview at more than fifty of its own repositories and watched it write, compile and run its own proofs of exploitability. Its refusals on identical code did not repeat.

Reality

Evidence34
Adoption26
Hype gap+18
Incentives62
Confidence44
securityOne report1 publisher

A machine found the bug and set the clock: Ledger disputes TestMachine's timeline

Ledger's CTO says the Ethereum clear-signing fix shipped two weeks before disclosure. The public release record cannot confirm that, and an autonomous scanner chose the publication date.

Reality

Evidence38
Adoption
Not scored
Hype gap+24
Incentives72
Confidence36