OpenAI says upcoming cyber model found 948 Linux kernel privilege-escalation flaws
Reported by Blockchain News, runtimewire.com, Unite.AI and 9 more outlets12 sources ↓
Latest in this story: OpenAI launches GPT-6 Cyber with 'Critical' cybersecurity rating
OpenAI says an unreleased cyber model identified 948 distinct Linux kernel flaws that could let users gain administrator rights. Its system generated patches, which human researchers reviewed before reporting findings to maintainers; some flaws have been fixed, while work on others continues.
Key points
- OpenAI says the system reproduced 1,724 vulnerabilities and developed working exploits for 948 distinct privilege-escalation flaws
- The Linux kernel analysis cost $1,213,430 in API credits, according to OpenAI
- Human researchers reviewed the generated exploits and patches before reporting findings to Linux maintainers
- TechRadar reports that 61 flaws had been patched and 43 had staging patches; other findings remain under review
Sources
- Blockchain NewsRead ↗OpenAI claims cyber model patched 948 Linux kernel LPEs
- runtimewire.comRead ↗OpenAI says its agents exploited and patched 948 Linux kernel flaws for $1.2M
- Unite.AIRead ↗OpenAI Pipeline Finds 948 Linux Kernel Privilege Escalation Flaws
- Startup FortuneRead ↗OpenAI's AI pipeline turned 2,308 Linux kernel bugs into 948 root exploits
- GBHackers NewsRead ↗OpenAI’s New Cyber AI Model Finds 948 Linux Privilege Escalation Vulnerabilities in Just 10 Days
Details
OpenAI says its agents scanned selected security-relevant parts of the Linux kernel, narrowed 2,308 candidates to 1,724 reproduced vulnerabilities, and developed working local privilege-escalation exploits for 948 distinct flaws. The system also generated proposed patches. OpenAI says human researchers reviewed the evidence and patches before contacting Linux maintainers.
The Decoder reports that OpenAI put the run’s compute cost at $1,213,430, or about $1,279 per exploit.
The findings are at different stages of remediation. TechRadar reports that 61 flaws had already been patched by the time the analysis ended, another 43 had staging patches, and some others were being publicly discussed; OpenAI says remaining findings are being addressed. The Decoder reports that 304 critical findings were sent to maintainers with suggested fixes.
Written from coverage as of
Images
- The Decoder
- runtimewire.com
- CoinCentral
- TechRadar