Cases › other › LR-2026-0727

XBOW autonomous agent finds Linux kernel DIBS loopback flaw enabling local root (CVE-2026-72018)

Preliminary Why: 1 report: Cyber Security News · subject to change · 1 independent report

Status: fix reported

Severity minor

Reported 30 Sep 2026

Developer and model XBOW

No source states: loss or records, agent tool, could do, deployer, industry.

Severity minor: the exploit work was contained to one experiment on Ubuntu 24.04 with kernel mitigations disabled, and no third party is stated as touched

XBOW, an autonomous security research platform, had its agent handle threat modeling, code auditing, bug discovery, validation and much of exploit development against the Linux kernel. The agent identified a missing bounds check in the DIBS loopback driver used by SMC-D, which allowed a copy beyond an allocated kernel buffer. The researchers built a local privilege-escalation exploit from a weak 16-byte zero write, and the reported proof of concept reached root on 22 of 100 boots in a test on Ubuntu 24.04 with mitigations disabled. The flaw was disclosed as CVE-2026-72018, and an upstream fix adds offset and size validation. Human researchers redirected the agent at several points. Administrators are advised to apply the kernel update and review CAP_NET_ADMIN grants.

What happened

1 · What it was doingXBOW's autonomous agent handled threat modeling, code auditing, bug discovery, validation, and much of exploit development against the Linux kernel.

2 · What went wrongA missing bounds check in the DIBS loopback driver let attacker-controlled data be copied beyond an allocated kernel buffer.

What was done

Administrators are advised to apply the kernel update with the bounds-check fix and reboot affected systems.

Fix reported

“Upstream remediation adds validation for offset and size before the copy operation.”

No later incident for XBOW under this risk on record.

Risks

AI used to run a cyber attack
Primary96 incidents
Rising: 24 vs 4.0 a week

Read from the sources; not yet reviewed by a person.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Model

Safety training and refusal · Cyber and weapons safeguards · Abuse monitoring and account enforcement · Staged or restricted release · Dangerous-capability testing before release · Outside red-teaming · Content safety classifiers · Safeguards kept on in testing · Reasoning monitoring

Infrastructure / target

Known flaws patched · Multi-factor sign-in and password policy · Network segmentation · Intrusion and endpoint detection · Backups and recovery · Phishing and fraud filters

Evidence

No source states: what it reached or did, 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.

Show the full account: 3 items from the primary sources

What happened, in order

Not dated in the sources:

The proof of concept succeeded on 22 of 100 boots, with the first escalation on the seventh boot.
Upstream remediation added offset and size validation before the copy operation.
Administrators were advised to apply the kernel updates and reboot affected systems.

Show every field, with its source and grade

The AI system

Developer

XBOW

C one secondary source states it