‘This work is a glimpse of what is coming’: Security team lays out how Anthropic Mythos helped build a working macOS exploit in five days

0 0 2 minutes read

‘This work is a glimpse of what is coming’: Security team lays out how Anthropic Mythos helped build a working macOS exploit in five days

https://www.profitableratecpm.com/f4ffsdxe?key=39b1ebce72f3758345b2155c98e6709c

California researchers used Anthropic’s Mythos Preview to chain two bugs and techniques into a macOS kernel exploit on Apple M5
The exploit bypassed Apple’s new memory integrity enforcement system, reaching the root shell in five days despite years of Apple investment.
The attack highlights Mythos’ power to reveal unknown flaws; Apple is reportedly working on a fix

Cybersecurity researchers from California have explained how they used Anthropic’s (now) famous Mythos Preview AI tool to create a working macOS kernel memory corruption exploit on Apple’s latest M5 silicon, warning that their work was a “preview of what’s to come” for hardware and software that was built “in a world before Mythos Preview.”

In September 2025, Apple introduced a new security feature designed to block hacking techniques that exploit software memory vulnerabilities.

Called Memory Integrity Enforcement (MIE), the feature uses hardware-level memory controls to prevent malicious code from accessing data it shouldn’t. It was released alongside the iPhone 17 range and the new A19 chips.

Execution model persists

In April 2026, Anthropic granted access to Mythos Preview to a handful of tech companies, including Calif, so they could get a head start on everyone else and secure their environments.

The company claimed that Mythos was capable of revealing unknown vulnerabilities and creating working exploits and as such it was too dangerous to simply make public.

Calif used Mythos to connect “two bugs and a handful of techniques to corrupt Mac memory and then access parts of the device that should be inaccessible.” Commenting on their findings, the California team said Apple spent five years and “probably billions” of dollars building MIE, while they managed to break it in five days.

“The exploit is a data-only kernel local privilege escalation chain targeting macOS 26.4.1 (25E253),” Calif said in a Substack article. “It starts with a local unprivileged user, uses only normal system calls, and ends with a root shell. The implementation path involves two vulnerabilities and several techniques, targeting bare-metal M5 hardware with the MIE kernel enabled.”

Apparently Apple is currently working on a fix. Probably using Mythos too.