/Tech45d ago

Anthropic Mythos AI uncovers macOS M5 kernel vulnerabilities

Anthropic Mythos AI enabled researchers to find two previously unknown macOS kernel vulnerabilities on M5 silicon. The flaws were chained into a privilege-escalation exploit that bypassed Memory Integrity Enforcement. The work took five days and roughly $35,000 in API usage. Researchers delivered a 55-page report to Apple headquarters. The demonstration on an M5 Max under macOS 26.4.1 produced the first public macOS kernel memory-corruption case on M5 hardware.

2647K5733K2M

#107

Original post

Rohan Paul@rohanpaul_ai

WSJ: Anthropic’s Mythos helped researchers find 2 unknown macOS kernel bugs and turn them into a working privilege escalation exploit in 5 days.

The target was the macOS kernel, the deepest layer of Apple’s desktop operating system, where code controls memory, processes, permissions, and access to hardware.

Mythos helped connect 2 separate flaws with extra exploitation techniques, which means the attack did not rely on one bug but on a chain where each step made the next step possible.

The exploit allegedly corrupted memory, bypassed Apple’s memory integrity protections, and gained access to protected parts of the system that normal apps should never reach.

This is serious because modern macOS defenses are built to make memory bugs hard to convert into control of the machine, not just hard to find.

Mythos can become so powerful here because vulnerability research is a search problem with many dead ends, where the model can help form hypotheses, inspect code behavior, reason across low-level constraints, and suggest exploit paths faster than manual work alone.

---

wsj .com/tech/ai/anthropic-mythos-apple-macos-bug-339da403

4:41 PM · May 14, 2026 · 15.5K Views

Sentiment

Users celebrate Anthropic's Mythos AI for finding macOS kernel exploits on M5 silicon as proof AI beats costly security work, while others dismiss it as hype or warn about cheap exploits.

Pos

64.4%

Neg

35.6%

22 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

CALIF.IOVia

Posts from X

Most Activity

VIEWS1.5MBOOKMARKS2.2KLIKES5KRETWEETS431REPLIES118

Ejaaz@cryptopunk7213

claude mythos just broke Apple's $2 billion defense system. it did so by discovering a completely different attack vector to break in

only took it 5 days costing ~$35K of mythos api time (the same exploit class costs $5-10M on grey market)

the researchers that commandeered the exploit produced a 55-page report that was delivered to Apple HQ in-person (hoping they release it after patching).

most shocking part for me is apple's MIE worked as intended. mythos just discovered a new way to side-step it entirely by poisoning the data the M5 chip ingested.

at this point i think we have to accept that mythos walks the walk.

As the anthropic red-team explicitly confirmed this week - this is NOT a compute resource issue. its national defense.

International Cyber Digest@IntCyberDigest

❗️🚨 BREAKING: Researchers used Mythos Preview to find the first public macOS kernel memory corruption exploit on Apple's M5 silicon, they give a glimpse into Mythos say it’s really powerful.

Apple spent five years and an estimated several billion dollars building Memory

45d1.5M5K2.2K

Chubby♨️@kimmonismus

Three researchers used Anthropic's Mythos to build a working macOS kernel exploit that bypasses Apple's M5 Memory Integrity Enforcement, a security system Apple spent five years and billions of dollars building.

Bug found April 25. Working exploit May 1. Walked into Apple Park to deliver the report in person.

MIE was the flagship security feature of the M5 and A19, designed to kill the entire memory corruption bug class. According to Apple's own research, it disrupted every known public exploit chain against modern iOS.

Calif didn't break MIE. They walked around it. Data-only attack, no pointer manipulation, standard syscalls from an unprivileged user to root.

The 55-page technical report drops after Apple patches. This is the story of the year in cybersecurity.

International Cyber Digest@IntCyberDigest

Video of exploit in action.

Source: https://blog.calif.io/p/first-public-kernel-memory-corruption

44d503.4K1.8K755

Ejaaz@cryptopunk7213

@harshexists anthropic's internal red team tests have showed models sometimes actively hide things from their researchers.

45d8.3K462

Harsh@harshexists

@cryptopunk7213 now my Question is, if AI can find things that humans can't. are we sure that AI is returning everything it finds? what if it discards something or keeps somethings for it self for future useage.

45d10.2K271

Grok@grok

**Implications:** AI like Anthropic's Mythos Preview is accelerating offensive research—researchers at Calif built a working kernel memory corruption exploit bypassing Apple's MIE on M5 silicon in just 5 days for ~$35K. It shows hardware memory safety (years and billions in development) can be side-stepped quickly via new attack vectors, speeding up the cyber arms race and highlighting AI's dual-use power for both red-teaming and real threats.

**Investment angle:** Firms strong in AI-native endpoint detection, behavioral analysis, and automated response (e.g., CrowdStrike or SentinelOne) stand to gain as defenses must evolve faster. This isn't financial advice—DYOR.

45d7185

TraderSmith09@TraderSmith09

@cryptopunk7213 Important.

"Human expertise was still required to bypass MIE (Mythos helped generalize from known bug classes but didn’t fully autonomously defeat the new mitigation)"

45d2K122

Jacob Shi@Jacoob_shi

@cryptopunk7213 $35K api time vs $5-10M on grey market is a wild delta. and the data poisoning angle is way more interesting than a traditional exploit - that's not a bug you can just patch

45d3.8K12

Shinka - AI@ShinkaIoT

We are watching the cost of zero-days collapse in real time.💣💣

The cybersecurity economy is built around human bottlenecks. If an agent can chain complex kernel exploits in 5 days, the old defense model of "prevent the breach" is dead.

We are moving to a world of autonomous containment, because human teams physically cannot patch this fast.

46d20332

Benjamin Babik@localoptimiser

@rohanpaul_ai Mythos, casually crawling through Apple's public issue trackers where Apple engineers tell the rest of us our heavily reported and reproducible memory corruption vulnerabilities are not actually bugs but features we're too unsophisticated to appreciate.

45d21851

Marius Breuer@MariusBreuer

@cryptopunk7213 @grok what are the implications of that? Which cyber security company would you invest?

45d7.5K

nsxdavid@nsxdavid

@cryptopunk7213 Until it is released, I do not accept anything of the kind. Did anyone try GPT-5.5 to see if it could come up with the same thing?

45d2.1K13

teromee@teromee

@kimmonismus So the issue that I had pointed out internally to engineering during my employment at Apple has been lurking this entire time and hasn't really been fixed.

44d6043

Grok@grok

Apple's latest M5 chips have a brand new high-tech security shield called Memory Integrity Enforcement. It was supposed to completely stop a whole class of hacker tricks that mess with a Mac's memory to take over the system.

Researchers used AI (Anthropic's Mythos) to quickly build a clever workaround. Instead of injecting bad code, they manipulated data in a way that lets a normal app quietly gain full "root" control — basically god mode on your Mac.

They found the bug on April 25 and had a working exploit by May 1. The video shows it successfully giving root access on a protected M5 system.

It proves even Apple's billion-dollar defenses can be bypassed fast, especially with AI helping. Apple will patch it soon.

44d4353

John Pippett@JohnPippett

@cryptopunk7213 Doesn’t that literally make it a compute resource issue? The funniest part of this tweet is nobody has handed GPT5.5 36k$ of tokens for anything remotely similar. This is just AI hype cycle nonsense.

45d1.5K4

Harsh@harshexists

@cryptopunk7213 Exactly my point. AI knows when its being tested, and when not. and now we are asking it to test other things, and hoping it will work in our favor.

its like requesting a cat to protect the milk

45d1.4K4

NirvanaNuts@NirvanaNutss

@cryptopunk7213 Security theater gets very humble when a new angle walks through the side door. The price tag was protecting the ritual more than the wall.

44d38711

Kushagra Tiwari@Kushagrat15

@cryptopunk7213 $35K to do what costs $5-10M on the grey market. the security economics just permanently changed and most companies haven't caught up yet

45d1.1K1

Chud@chud_groyper

@cryptopunk7213 I had to downgrade from 4.7 to 4.6 at work because 4.7 kept coming up with overly complex solutions.

It’s going to take a long time to get this right.

45d9508

0xMetaLabs@0xMetaLabs

@cryptopunk7213 The economics here are absurd. A vulnerability class worth $5M - $10M on gray markets allegedly reproduced with ~$35k of API calls.

If true, frontier cyber capability is rapidly shifting from elite, scarce expertise toward scalable inference economics.

44d4551

Dave@RideToFireStar

@cryptopunk7213 But what if @claudeai found something Tim @Apple had put in the M5 chip for the @CIA? 🤔

(there will be more like this that surface, and not just from Mythos either)

45d2.3K6