OpenAI says an autonomous AI pipeline built working root exploits for 948 unique Linux kernel vulnerabilities in one run, a result that puts the phrase Agents Got Root at the center of a new security challenge, according to SecurityWeek’s report. The company said the scan cost $1.2 million in API credits and targeted the latest code in the kernel’s master branch in mid-October.
The pipeline used an unreleased cyber model. Its agents narrowed the kernel to areas reachable by low-privileged users, flagged 2,308 potential bugs and confirmed 1,724. They tested their exploits on virtual machines running an unmodified desktop Linux setup with common protections enabled, including KASLR, SMEP and SMAP.
Agents Got Root Across 948 Separate Patches
OpenAI reported 1,169 working local privilege escalation exploits and 291 proofs of concept for information leaks. It said the 948 figure counts separate patches associated with the bugs behind the exploits. Before sending material to kernel maintainers, OpenAI security researchers reviewed the exploits, proposed fixes and severity assessments.
“To the best of our knowledge, the agents in this pipeline discovered and used at least 10 novel Linux exploitation techniques,” the company said. OpenAI added that outside kernel exploitation experts were not familiar with those techniques. It is withholding full technical details for safety reasons.
The scan’s bill was $1,213,430. Spread across 948 exploits, that works out to roughly $1,279 apiece. An earlier scan in late July, powered by GPT-5.6 Sol, produced 166 unique local privilege escalations at a cost of $513,690.
KernelCTF Run Produced 21 Exploits
OpenAI compared the scan’s results with historical prize amounts. The company said Pwn2Own and TyphoonPWN had paid $20,000 and $70,000 for Linux kernel privilege escalation exploits. At those amounts, it valued the 948 bugs at $18.9 million to $66.3 million, or 16 to 55 times the scan’s cost.
OpenAI said the October scan found more than five times as many new local privilege escalations as the July scan, despite three months of hardening in between. It also tested the pipeline against an adapted version of Google’s kernelCTF, which targets a hardened kernel with less exposed attack surface. Over 112 hours and about $33,000 in API credits, the run delivered 21 local privilege escalation exploits.
Hundreds of reports and patches for critical flaws have already been submitted, OpenAI said, following advice from the kernel community. The company plans to send fixes for the remaining bugs in stages with partners in its Patch the Planet initiative. OpenAI also reported that within 10 days of the scan starting, more than 100 Linux local privilege escalations had been patched or had staging patches from other contributors, while about 300 more were publicly discussed.
OpenAI Urges Layered Defenses
OpenAI warned that a low-privileged account on an operating system missing its latest available patches should be treated as functionally equivalent to root access for a well-resourced threat actor, at least in the near and medium term. The company pointed to sandboxing and isolation technologies, including Firecracker and gVisor, as ways to contain damage if a flaw is exploited.
The source also described a separate OpenAI announcement on November 2 about higher Security Bug Bounty payouts. OpenAI’s earlier security work is also relevant: its cybersecurity model drew attention from defenders. OpenAI said the midpoint of its payout ranges for medium-, high- and critical-severity bugs is now more than four times higher on average. Reports of critical flaws in ChatGPT, Codex, the API Platform or other core products can earn as much as $30,000; the exceptional critical-bug cap remains $100,000, while bonuses can raise a single payout to $172,000.
The company is adding bonuses for high-quality reports, a five-tier loyalty program for researchers who find valid bugs, and extra multipliers for flaws reported during priority campaigns. Its next stated step on the kernel findings is to continue sending fixes in stages with Patch the Planet partners.
