Skip to content
MaungaAI news, explained

Safety and security

OpenAI AI finds 948 Linux flaws that hand over control of a computer

OpenAI says one of its unreleased models found 948 flaws in the Linux kernel and proposed fixes for each one. The flaws give full control of the computer. The test cost $1.2 million, and so far only the company is reporting it.

2 min read2 sources, one officialTranslated with AI from the Spanish original

Share: WhatsApp · Email

AI-generated illustration for Maunga. Not a photograph.

The essentials

  • Claimed by one partyAccording to OpenAI, one of its unreleased models found 948 flaws in the Linux kernel that allow someone to take full control of the computer (OpenAI News, ciberseguridad).
  • Claimed by one partyOpenAI says the test cost $1,213,430 in use of its AI, about $1,279 for each flaw exploited (ciberseguridad).
  • Claimed by one partyAccording to OpenAI, company researchers reviewed each flaw and its fix before sending them to the people who maintain Linux (ciberseguridad).
  • Claimed by one partyOpenAI says its agents used at least 10 new attack techniques. It is keeping the details secret for security reasons (ciberseguridad).

Why it matters

If the figures are confirmed, finding serious flaws in Linux has become fast and cheap. That helps defenders, but it also helps attackers. The UK's cybersecurity centre, the NCSC, believes it is almost certain that AI tools have sharply increased the number of flaws discovered (ciberseguridad). This suggests the challenge is no longer so much finding flaws as applying the fixes in time.

What does it mean for me?

If you use Linux or manage servers, updating can't wait. OpenAI warns that on a system without the latest fixes, an account with no special permissions is almost the same as full control for a well-resourced attacker (ciberseguridad).

The details

The test was run in mid-October 2026 against the latest code of the kernel, the central part of Linux (ciberseguridad). According to OpenAI, its agents, AI programs that work on their own, flagged 2,308 possible flaws and reproduced 1,724 of them in an isolated environment (ciberseguridad). They then built exploits, which are programs that take advantage of a flaw, and wrote a fix for each one. Company researchers reviewed everything before alerting the people who maintain Linux (ciberseguridad).

The sources do not give the same figure. One outlet reports 948 exploits (ciberseguridad). Another says the agents produced 1,169 exploits and that 948 is the number of distinct fixes needed (ciberseguridad). According to OpenAI, it has already sent hundreds of reports and fixes for the critical flaws, and will send the rest in stages (ciberseguridad).

What we don't know

  • Whether anyone outside OpenAI has checked the flaws and the figures from the test.
  • When OpenAI will release the model it used, and who will be able to use it.
  • How many of the fixes the Linux kernel maintainers have already accepted.

How this story has changed

  1. There are now details on the method, the cost, the human review and OpenAI's warning about systems that have not been updated, as well as a discrepancy between 948 and 1,169 exploits.
  2. We published the story.

Sources

This story is based on 3 sources. In the text, each fact names in brackets who reports it.

Was this story useful? Yes Not really

Spanish original