Severity negligible: Nothing beyond the system itself was affected and nothing was left changed. The exploits were lab proof-of-concepts never seen in a real attack, and every flaw was reported to the vendor and patched.
Earliest source we hold 2026-10-28 · Captured by AI Risk Wire 2026-10-28 16:11 UTC
Anthropic's Frontier Red Team reported that an unreleased internal model, prompted with little human involvement, found flaws in Chrome, Safari, Android, Windows and Linux and chained them into working proof-of-concept exploits in the lab.
Risk grouping: Not yet classified
No source we hold states: who caused it, intended or not, before or after release, the ai's part.
The story
1 · What it was doingAnthropic's Frontier Red Team prompted an unreleased internal model to hunt for security flaws in Chrome, Safari, Android, Windows and Linux and chain them into working exploits, in the lab rather than against real targets.
2 · What went wrongThe model produced working proof-of-concept exploits, including a Chrome chain escaping the browser sandbox and a Linux chain giving root over the network, capabilities the company says were previously the work of specialist teams.
3 · What it reached or didThe chains reached Chrome on macOS, Safari on iOS and iPadOS, a locked Android phone over USB, Windows servers answering HTTP/3, and Linux servers running a multipath TCP listener. No real attack occurred.
4 · What was doneAnthropic reported each flaw under a coordinated disclosure policy giving developers 90 days to patch and 45 more before details beyond a public advisory; fixes landed in Chrome 156, iOS/iPadOS 26.5.2, CVE-2026-33096 and CVE-2026-98432.
No source states: how it was caught, when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, the cause, what set it off.
D inferred by us from the described behaviour, reason shown
Stage of use
in research
C one secondary source states it
Whose system
company that built the AI
D inferred by us from the described behaviour, reason shown
Stage of the model
before release
C one secondary source states it
What it was for
Security research: finding and chaining software vulnerabilities into proof-of-concept exploits
C one secondary source states it
What happened
Earliest source we hold
2026-10-28
A the date the source itself carries
Near miss
yes, no harm happened
C one secondary source states it
What was done
Coordinated disclosure with 90 days to patch and 45 more before details beyond a public advisory; fixes landed in Chrome 156, iOS/iPadOS 26.5.2, CVE-2026-33096 and CVE-2026-98432.
D inferred by us from the described behaviour, reason shown
What went wrong
other
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
C one secondary source states it
Impact
Severity
negligible
D inferred by us from the described behaviour, reason shown
Kind of harm
no harm happened
D inferred by us from the described behaviour, reason shown
Why this severity
Nothing beyond the system itself was affected and nothing was left changed. The exploits were lab proof-of-concepts never seen in a real attack, and every flaw was reported to the vendor and patched.
D inferred by us from the described behaviour, reason shown
LR-2026-0265 · version 2 · published 2026-10-28 after automatic checks; not yet checked field by field by a person · changes to this case · JSON