Oh, so you thought Mythos and other AI models would only find vulnerabilities in software? Well, finding weaknesses in firmware and hardware is traditionally much more difficult, requiring specialized skills and in some cases very expensive tools. But AI is very proficient at specific types of hacks like memory corruption. Such architectures are embedded in CPUs, GPUs, and other processors as well as system memory.

This week, vulnerability researchers used Mythos to identify and exploit a kernel memory vulnerability in the Apple M5 computer chip that powers MacBooks and iPads. That is the core brain of the system.

The strategic takeaway is that new AI models like Anthropic’s Mythos will greatly expand the capabilities of responsible and malicious vulnerability researchers to find weaknesses in technology throughout the stack. Data, applications, interfaces, operating systems, virtualization environments, network transports, firmware, and hardware systems are all on the menu when it comes to the next generation of vulnerability detection and exploit creation.

Full Article: https://open.substack.com/pub/matthewrosenquist/p/cybersecurity-must-prepare-for-ai