Anthropic’s AI model called Mythos Preview can analyze security patches for Firefox and Windows and turn them into working cyberattacks within hours. The company’s security team tested this and found eight complete attack methods were ready before Microsoft’s automatic updates even reached a single device.
This is scary because security patches are supposed to protect us. When companies like Microsoft or Mozilla find a security hole, they release a patch to fix it. But now AI can look at that patch, figure out exactly what the hole was, and create an attack to exploit it faster than people can install the fix.
The Old Security Game Is Broken
The whole system used to work because hackers needed weeks or months to figure out how to exploit security patches. That gave most people time to install updates. But Anthropic’s AI did it for just a few thousand dollars with no special hacking skills required.
The AI created working exploits while regular users were still waiting for their computers to automatically download and install the security updates. It’s like someone publishing the answer key to a test before students even get their test papers.
What Happens Next
Anthropic says this proves the current patch system is “obsolete.” Companies might need to change how they release security fixes, or find ways to push updates much faster. Some experts think this could force a complete rethink of cybersecurity.
The bigger worry is what happens when this AI technology gets into the wrong hands.
