The AI Read is on a break. The last edition was November 15, 2026. Everything here stays open to read.
The AI Read
← AI News

Illustration, AI-generated from the story text. Tap a panel to read the story through it.

Morning Brief · November 5th 2026

OpenAI says one automated run produced 948 working routes to root on Linux, at $1,279 each

OpenAI published the results of a pipeline that used an unreleased "upcoming cyber model" to find, exploit and patch flaws in the Linux kernel. In a mid-October run against about 2.8 million lines, roughly 7% of the kernel, it produced 2,308 deduplicated findings in 12 hours and reproduced 1,724 in 30 hours. It built 1,169 working local privilege escalation exploits and 291 information-leak proofs, tested in virtual machines with standard defenses switched on, and counts 948 as unique because that is how many distinct patches they needed.

The run cost $1.213 million in API credits. A late-July run with GPT-5.6 Sol cost $513,690 for 166 unique exploits, so the price per exploit fell from about $3,095 to $1,279 in three months.

Read the patch count carefully. All 948 are candidates for maintainers to review, and none is a merged fix. By the end of the scan 61 of the flaws were already patched upstream, 43 had patches in staging and 299 had been publicly reported or discussed. OpenAI says it has sent hundreds of reports and will release the rest gradually with its Patch the Planet partners, with humans reviewing each one first. Gizmodo's $18.9 million to $66.3 million value assumes every exploit would earn a bounty, which nobody has offered. SourcesABC

Sources

This story ran in the Morning Brief of November 5th 2026.