OpenAI's AI Finds 948 Security Flaws for $1.2 Million
This could make your phone and the internet safer—but at a steep price.
OpenAI recently revealed that it used a powerful AI system to automatically hunt for security bugs in the Linux kernel—the foundational code that powers Android phones and most of the internet. The AI found over 2,300 potential vulnerabilities, confirmed 948 were truly dangerous, and even wrote patches (fixes) for them. The entire process cost $1.2 million in AI tokens (the units of text the AI processes), which sounds like a lot, but OpenAI argues it's a bargain compared to paying human hackers.
Why should you care? Security flaws in the Linux kernel affect billions of devices, from your smartphone to the servers that run your favorite websites. Finding and fixing them quickly makes the digital world safer for everyone. Traditionally, companies pay independent hackers $20,000 to $70,000 per bug through 'bug bounty' programs. OpenAI's AI did it for about $1,279 per validated flaw—a potential savings of 16 to 55 times. That could mean faster fixes and fewer vulnerabilities that criminals can exploit.
The catch is that the AI doesn't actually apply the fixes; it just suggests them. Human experts still need to review and approve each patch, which takes time and expertise. Also, the $1.2 million price tag is just for one run—and that's the token cost, not the total project expense. So while the AI is impressive, it's not a fully automated solution yet. It's more like a super-powered assistant that speeds up the work of human security teams.
This isn't OpenAI's first foray into vulnerability discovery; they did a smaller version in June called 'Patch the Planet.' Google also has a similar tool called Big Sleep. The trend is clear: AI is becoming a key player in cybersecurity, potentially making our software safer and saving companies money. But for now, it's a tool that augments human efforts, not replaces them.
- OpenAI's AI found 948 real security flaws in Linux, the software behind Android and the internet.
- The AI cost $1.2 million to run, but that's only $1,279 per flaw—much cheaper than human hackers.
- Humans still need to approve the fixes, so it's not fully automatic yet.
Why It Matters
Faster, cheaper bug fixes could make your devices and online services more secure.