OpenAI produced 948 root exploits on Linux for 1.2 million dollars

On November 4, 2026, OpenAI published the results of an agent pipeline that reproduced 1,724 Linux kernel vulnerabilities and wrote 948 privilege escalation exploits, at 1,279 dollars each. The next day, GPT-6 Cyber was classified as "critical" and restricted to verified defenders. That same week, Moonshot removed cybersecurity from the training of its open model.

In short

• On November 4, 2026, OpenAI published the results of an agent campaign against the Linux kernel: 948 validated root exploits for 1.2 million dollars in API credits, or 1,279 dollars per exploit.
• A five-stage pipeline; writing the exploits is the most expensive stage, 775,929 dollars. A check was added after 20 exploits out of 200 had obtained root through a different flaw than the one targeted.
• The next day, GPT-6 Cyber, rated "critical" in cybersecurity, reserved for the Daybreak Red program; it also found 81 unknown high-severity flaws in Firefox.
• The same week, Moonshot released Kimi K3.1 Pro after removing cybersecurity data from training, from 90.5% to 81.0% on its benchmark. Two laboratories, one choice: offensive capability stays with whoever produced it.

← All the watch