OpenAI’s Frontier Cybersecurity Model Finds About 1,000 Linux Local Privilege Escalation Vulnerabilities in One Week
additional context on the project: https://x.com/Eric_Wallace_/status/2118123944251453949
AI BriefingOpenAI is applying substantial frontier compute to freely discover and fix vulnerabilities in open-source code repositories. Its upcoming frontier cybersecurity model found approximately 1,000 local privilege escalation (LPE) vulnerabilities in the Linux kernel within one week, at a cost of approximately $1M in tokens, and discovered at least 10 new categories of Linux exploitation techniques. By comparison, security researchers previously typically found around 20 LPE vulnerabilities in Linux each year. Greg Brockman reposted the project with additional details, saying that the Linux kernel is especially critical software for OpenAI.
Also covered by 7 other sources
- The OpenAI team used a single run of a cyber model to validate and patch 948 local privilege-escalation vulnerabilities in the Linux kernelTibo · 12 days ago
- OpenAI Says Its Security Model Fixed 948 Linux Kernel Privilege Escalation Vulnerabilities in a Single Run, at an API Cost of About $1.2 MillionRohan Paul · 12 days ago
- OpenAI Finds and Fixes 948 Linux Kernel Local Privilege Escalation Vulnerabilities with a Cyber Vulnerability PipelineGreg Brockman · 12 days ago
- OpenAI Demonstrates an Autonomous Vulnerability Discovery and Patching Pipeline, Generating Patches for 948 Linux Kernel Local Privilege Escalation Vulnerabilities in One RunOpenAI · 12 days ago
- OpenAI Demonstrates a Model Reproducing 1,724 Vulnerabilities and Generating Exploits for 948 Local Privilege Escalation VulnerabilitiesOpenAI · 12 days ago
- OpenAI reveals the cost of using AI to uncover local privilege-escalation vulnerabilities: about $1,279 per vulnerabilityOpenAI · 12 days ago
- OpenAI Uses a Cyber Model to Automate the Discovery and Patching of 948 Linux Kernel Local Privilege Escalation VulnerabilitiesOpenAI: Official Website Updates · 12 days ago
Source: Greg Brockman · x.com
Event follow-up3
View all event coverage- OpenAI releases GPT-6 Cyber cybersecurity modelOpenAI: Official Website Updates · 11 days ago
- OpenAI Says Its Agents Discovered at Least 10 New Linux Exploit TechniquesOpenAI · 12 days ago
- OpenAI Uses a Cyber Model to Automate the Discovery and Patching of 948 Linux Kernel Local Privilege Escalation VulnerabilitiesOpenAI: Official Website Updates · 12 days ago