Skip to content
AgentProv

2026-09-28 · 2026-Q3

XBOW's AI Agent Finds and Exploits a Linux Kernel Out-of-Bounds Write to Root (CVE-2026-72018)

Outside the model: AI capabilityAgent tooling was not the attack surface, so no layer applies.

Fixable
Vendor patch
Vendor response
Patched
In the wild
No
Actor
Not applicable

Target

The Linux kernel's SMC-D (Shared Memory Communications) implementation, `dibs_loopback` driver, in recent kernels with DIBS loopback support

Impact

XBOW published on September 28, 2026 how its autonomous system did the bulk of the threat modeling, code auditing, discovery, and validation for a 16-byte zero-write at an attacker-controlled offset in kernel heap memory, then built a working local privilege escalation from that constrained primitive alone by targeting the identity fields of the kernel `cred` structure, with no separate memory leak. The exploit needs `CAP_NET_ADMIN` and succeeded in 22% of 100 boots on Ubuntu 24.04 (Linux 7.1.0-rc6). Three human interventions redirected the agent's reasoning during the campaign. Reported June 19, patched July 7, CVE assigned August 17

Root cause

A kernel memory-safety bug in a new subsystem, found and weaponized by an agent at a pace that compresses the window between a fix landing and a working exploit existing

CVE

Sources

Catalog

Cite this: AgentProv incident record: XBOW's AI Agent Finds and Exploits a Linux Kernel Out-of-Bounds Write to Root (CVE-2026-72018) (2026-09-28). https://agentprov.com/incidents/2026-09-28-xbow-s-ai-agent-finds-and-exploits-a-linux-kernel-out-of-bounds-write-to-root