OpenAI's AI Uncovers 948 Linux Kernel Security Flaws
OpenAI says its automated system scanned the Linux kernel and found 948 ways to gain higher system access, spending $1.2 million (USD) in compute credits. For anyone trusting a phone, wallet app, or smart device built on Linux, this shows how fast AI now hunts for flaws.
What actually happened
According to Decrypt, OpenAI released its results on Wednesday. Its agents reviewed about 2.8 million lines of kernel code, near 7% of the project. They flagged 2,308 possible bugs, confirmed 1,724, and built 948 working privilege-escalation exploits. The October run cost $1,213,430 in API credits, about $1,279 per flaw. OpenAI compared that to the $20,000 to $70,000 paid for similar bugs at hacking contests, implying a manual value of $18.9 million to $66.3 million. An earlier model, GPT-5.6 Sol, found only 166 flaws in July for $513,690. The newer, unreleased October model found over five times more, using at least 10 attack methods unfamiliar even to OpenAI's own researchers.
How we got here
Linux runs most internet servers and a large share of phones and wearables. OpenAI chose the kernel to test how far automated bug hunting scales against code this large, per Decrypt. Days earlier, Anthropic said an unreleased Claude model had found and exploited serious Chrome and Linux holes, some allowing remote takeover with little user action. Separately, Vidoc Security rebuilt similar findings tied to Anthropic's Mythos AI for under $30 per file using public models. Both cases show AI bug hunting moving from slow manual work to cheap automation within months.
Why this matters for you
For wallet holders, the risk stays low today, since these exploits need local device access first, not remote takeover. For builders of connected hardware, including Linux-based wearables and smart glasses, faster bug discovery means patches must ship faster too. OpenAI limits its strongest tools to vetted defenders through a program called Daybreak Red, but warns attackers with partial access to old systems could use similar AI to escalate privileges faster. The simple takeaway for everyday users: keep devices and wallet apps updated, since patch speed now matters more.
The bigger question
If AI can find and weaponize hundreds of flaws in one of the world's most audited codebases for about $1,279 each, how long can defenders stay ahead of attackers using the same tools? The outcome may depend less on the technology itself and more on who controls access to it first.
What to watch
OpenAI shared its Linux findings on Wednesday but gave no release date for the newer cyber model used in October. Its Daybreak Red program keeps adding vetted defenders. Watch for how Linux maintainers respond, and whether Anthropic or other labs publish their own exploit counts soon.
