Anthropic’s specialized Claude Mythos AI model helped a team of researchers uncover a critical vulnerability in Apple’s systems. The version of the AI, which the company is currently making available only to select partners, allowed researchers to bypass macOS kernel memory protection on the latest M5 processors.
Anthropic’s specialized Claude Mythos AI model helped a team of researchers uncover a critical vulnerability in Apple’s systems. The version of the AI, which the company is currently making available only to select partners, allowed researchers to bypass macOS kernel memory protection on the latest M5 processors.
According to The Wall Street Journal, the first results of Claude Mythos Preview were demonstrated by cybersecurity experts from the Calif. company. The researchers used the Anthropic toolkit to find errors in the code of Apple’s current operating system.
Calif’s blog called the incident «the first public exploit to cause macOS kernel memory corruption on an Apple M5.» According to technical experts, the vulnerability allows a local user without high privileges to gain full access to the device. This was made possible by a combination of two bugs and several specific hacking techniques.
Calif representatives emphasized that AI played a key role in developing the exploit.
«Mythos Preview is powerful: once it has learned to attack a particular class of problems, it generalizes to almost any problem in that class. Mythos quickly detected these bugs because they belong to known classes,» the company noted.
The status of the vulnerability patch remains unclear. MacRumors spotted Calif and Anthropic Research mentioned in the release notes for macOS Tahoe 26.5, which was released last Monday. However, the researchers themselves said they only met with Apple earlier this week and will not release full technical details until the manufacturer has finally eliminated the attack vector.
An Apple representative in a comment to the WSJ limited himself to the standard statement that user security is the company’s priority, and all reports of vulnerabilities are carefully investigated.
As a reminder, Anthropic decided not to release Claude Mythos to the general public due to its extremely high efficiency in finding security holes. Currently, only large technology corporations such as Apple and a limited number of research organizations have access to the model to prevent the technology from being used by attackers.
Previously, Nicolas Carlini, an artificial intelligence researcher associated with Anthropic, said that the company’s latest model , Claude Mythos, quickly bypassed security protocols and gained access to confidential data.
