AICybersecurityNewsOpenAI

OpenAI’s AI Agents Found 948 Ways to Root Linux — the Same Week Voters Rejected Data Centers Almost Everywhere

Two AI stories landed on top of each other this week, and they pull in opposite directions. On November 4, OpenAI said its autonomous agents found and wrote exploits for 948 ways to get root on Linux. The day before, voters in local races across the country turned down AI data centers in nearly every ballot measure that asked about them. One story shows how fast AI capability is moving. The other shows the communities that have to host it pushing back.

948 paths to root, for $1.2 million in API calls

In a research post titled “A thousand fewer paths to root”, OpenAI describes an autonomous pipeline built on an upcoming, unreleased cyber model. In a single run in mid-October, the agents:

  • analyzed 2,308 candidate findings in the Linux kernel
  • reproduced 1,724 of them as real vulnerabilities
  • built 1,169 working local privilege escalation exploits, which let an ordinary local user take full root control, plus 291 information-leak proofs of concept
  • boiled those exploits down to 948 distinct vulnerabilities, and generated a candidate patch for each

The API cost for the whole run was $1.2 million. OpenAI calls that an upper bound and says better tooling should bring it down. For comparison, OpenAI says historical bug bounty payouts would value those 948 exploits at somewhere between $18.9 million and $66.3 million, roughly 16 to 55 times what the run cost. It also says the agents found and used at least 10 Linux exploitation techniques that are new, as far as it knows.

Some context matters here. “Patches” means candidate fixes that human researchers review and send to the kernel maintainers. They are not fixes that already shipped. OpenAI says that during 10 days of the scan, more than 100 of these local privilege escalation bugs were patched or had staged fixes from other contributors, and about 300 more were being discussed publicly. These are local bugs, so an attacker needs some kind of access to the machine first. Still, OpenAI’s own warning is blunt: for now, a low-privilege account on a Linux system that isn’t fully patched should be treated as “functionally equivalent to root access” for a well-resourced attacker. That system could be a server, a cloud host or an Android phone.

Meanwhile, at the ballot box

On November 3, data centers were on the ballot across the country, and they mostly lost. CNBC counted 41 data center measures in 10 states, and voters sided against development in every one except Harwood, North Dakota. Tallies differ by outlet: The Washington Post counted at least 38 places that rejected or restricted them. The measures ranged from outright bans and moratoriums to energy caps and requirements for a public vote.

Two days later, on November 5, OpenAI published a National Community Compact for its future data center campuses. It promises to pay the full cost of the electricity it uses and the grid infrastructure needed to serve it. It also promises site-level reports on water and power use, and an end to non-disclosure agreements with public officials, with narrow exceptions for security information and trade secrets. OpenAI doesn’t mention the election in the post, but the timing is hard to miss. One limit matters: the compact covers only campuses OpenAI owns and operates, not the leased or partner-built capacity behind much of today’s AI compute.

PCGeek’s take

These two stories are really one story. Results like the Linux run are the reason AI companies want ever-bigger compute campuses: the models keep getting more capable, and capability takes power. But the neighbors are now voting on whether they want those campuses at all, and they mostly said no. The useful part for most readers is the security one. If AI can turn a 7-figure API bill into nearly a thousand root exploits, the time between “bug found” and “bug exploited” is shrinking fast. Keep your Linux systems, servers and Android devices updated, and take kernel updates seriously.

Which story worries you more: AI that finds a thousand root bugs, or the fight over where its power comes from? Tell us in the comments.

Follow @pcgeekinfo for more.

PCgeek

Techie, YouTuber, Writer, Creator

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.