First Apple M5 memory exploit discovered using Anthropic AI

1

Join the conversation

Follow us

Add us as a preferred source on Google

Thanks to AI-assisted security research, hackers with hats of various colors are finding exploits everywhere. Linux has had its worst week in years with the CopyFail and Dirty Frag root-gaining vulnerabilities, and things aren't much rosier at Microsoft, thanks to the YellowKey BitLocker bypass, as well as GreenPlasma and RedSun privilege-gaining exploits. Now, it's Apple's turn with a local privilege escalation that gets past the M5 chips' much-vaunted Memory Integrity Enforcement (MIE).

There aren't many technical details, but the vulnerability is simple in practice: run a command as a standard user and gain root (administrator) access to the machine. Macs are rarely servers, so the practical impact is limited. However, the exploit remains concerning, as it's relatively easy to trick a user into running it and, with full system control, also hard to find and remove. The research team in question is named Calif, and as far as they know, the boffins there are the only ones making a public disclosure of this issue. Such assumptions are tricky in this day and age, though.

Mercifully for Captain Cook's ship, instead of being a zero-day reveal out of nowhere that left systems administrators scrambling, the exploit in question was disclosed to the company in advance (in person, no less). Calif published the vulnerability overview as part of a series of blog posts called the Month of AI-Discovered Bugs, since this new Apple vulnerability falls within a set of security findings aided by AI tools — in this case, Anthropic's Mythos Preview.

Latest Videos From

Go deeper with TH Premium: AI and data centers

(Image credit: Microsoft)

Photonics and high-speed data movement is the next big AI bottleneck

The data center cooling state of play

Massive AI data center buildouts are squeezing energy supplies

Ultra Ethernet: The data center interconnection of tomorrow

The researchers tested their code on an Apple M5 machine and macOS 26.4.1. The exploit chain impressively sneaks past MIE, a security feature present on M5 and A19 chips that labels each 16-byte memory slice with a 4-bit tag associated with the pointers that use it. MIE is enforced at the hardware level in a hypervisor-like configuration and effectively protects against most common classes of security exploits, namely, but not only, buffer overflows and use-after-free vulnerabilities.

As an oversimplification, MIE ensures that any memory read or write operation acts on the data that it was originally meant to, even at the kernel level. If that doesn't happen, either your application has a bug, or someone's up to shenanigans. The base feature is part of ARM MTE, and MIE is an Apple-added layer that enforces the said checks at the hardware level, with purportedly little to no performance overhead, and only 3% memory wastage. This blog post goes into more detail on the subject, and it's quite an interesting read.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

TOPICS

AI-assisted security research is producing exploits at a frightening rate.

Not really frightening. I think its good overall that these things get discovered in the glow of sunlight instead of some bad actor discovering it in the dark and then years go by before there is a fix but in the mean time they mine people's data unknowingly.

All software and hardware has exploits we just have not yet discovered them.

Reply