Anthropic’s Mythos AI Discovers First MacOS Exploit on M5 Chips

It took Apple five years to make it practically impossible to execute an exploit on its M5 chips, and now the first vulnerability has been made public by security researchers from a Palo Alto company called Calif, who have been able to find it. in five days with the help of Mythos PreviewAnthropic’s next AI.

Calif states from its website that Mythos Preview, Anthropic’s next generation of generative AI known for its security capabilities, has been key in identifying a memory corruption exploit of the macOS kernel on the M5 chip.

They did not want to give details about how the ‘exploit’ worked. until Apple fixes the vulnerabilities and the attack path used to penetrate the M5 chip. But they have explained that they worked with Mythos to identify the vulnerabilities, to help them with the development of the code for the ‘exploit’.

Mythos Preview was able to identify bugs quickly because they belonged to known classes. That does not mean that the developers gave the final touches to the design of the codealthough they make it clear that the help of Mythos AI has been essential in establishing patterns and finding errors that were previously completely unknown, and that could have been used by malicious actors to cause security breaches in Apple’s system.

Calif researchers claim that with the help of Mythos Preview they were able to create the exploit code in just five days. The ‘exploit’ is a data-only local kernel privilege escalation chain, targeting macOS 26.4.1. Start from a local unprivileged user, use only normal system calls, and end with a root shell.

Mythos Preview helped identify errors and assisted in the development of the exploit, But it not only learned how to attack one class of problems, but also generalized to almost any problem of the same class, as the researchers explain.

The fact that Calif has been able to design an exploit in a system that Apple spent five years creating to avoid memory corruption vulnerabilities, highlights the use of advanced AI systems capable of unraveling vulnerabilities in record time.

“Security is our top priority and we take reports of potential vulnerabilities very seriously,” Apple told The Wall Street Journal. Researchers have already met with the company at Apple Park in Cupertino to discuss what they are calling the “first public macOS kernel memory corruption exploit on M5 chips.”