A security startup claims to have achieved what many considered almost impossible: breaking the new defenses of the Apple M5 chip in just six days, with the help of Claude MythosAnthropic’s artificial intelligence model that is not available to the general public.
What exactly Anthropic’s AI did with the M5
The Vietnamese company Calif claims to have developed the first public macOS kernel memory corruption exploit capable of overcoming an Apple technology called Memory Integrity Enforcement (MIE), integrated into the M5 chips.
Work began on April 25. Six days later they already had a functional exploit that allowed them to obtain root access from an unprivileged native account. To achieve this, they combined two different vulnerabilities along with techniques specific to the M5 hardware and the kernel’s MIE defenses.
According to Calif, Claude Mythos was actively involved in identifying vulnerabilities and in different phases of the exploit’s development, although they clarified that human researchers were still necessary to overcome the most complex security barriers.
“Achieving a kernel memory corruption exploit against the best protections in a week is famous,” the company wrote. “It demonstrates what happens when advanced AI models are combined with security experts.”
Why you can’t use Claude Mythos yet
Claude Mythos is not just another chatbot. Anthropic presented it in April 2026 as an experimental model specialized in offensive cybersecurity and advanced vulnerability analysis, and from the beginning limited its access to a select group of companies within a private program called Challenge Glasswing.
Only selected technology companies, banks and researchers can use it. The results that are already known are striking:
- Identified 271 vulnerabilities in Firefox during internal tests reported by Mozilla
- UK AI Security Institute confirmed it can run sophisticated multi-stage cyberattack simulations
- US government agencies, including the NSAthey would have used the model for cybersecurity investigations
What we see here is an AI model that operates in a different category, not only more powerful than conventional models, but with capabilities that generate real debate about its responsible use.
Why this case changes the rules of cybersecurity
The memory corruption attacks They are among the most serious threats to any modern operating system. They allow everything from the theft of information to total control of the device. Apple designed MIE precisely to make them extremely difficult to run.
The problem is that MIE was designed in a pre-Mythos ecosystem. “Apple built MIE in a world before Mythos Preview,” Calif pointed out directly. “Now we will see how the best mitigation technologies hold up against the first big AI-driven ‘bugmageddon’.”
So far, Apple has not issued any public comment on the report or related security updates. That, in itself, says something. Specialists agree that the balance between attackers and defenders is about to break in the desire of those who use offensive AI. And the case of the M5 is the most concrete evidence that exists so far.
What is Memory Integrity Enforcement (MIE) on the M5 chip?
It is a security technology integrated by Apple in its M5 processors that uses advanced memory tagging and validation mechanisms to make it difficult to exploit kernel memory errors. It was designed to significantly raise the barrier against memory corruption attacks.
Can anyone use Claude Mythos to hack devices?
No. Claude Mythos is not available to the public frequently. Anthropic limits it to selected technology companies, banks and researchers within the private Challenge Glasswing program. Its access is controlled precisely by the level of risk it represents.
Has Apple already fixed the vulnerability in the M5 chip?
To date, Apple has not issued any public comments on the Calif report nor announced security updates related to this exploit. The company was notified directly during an in-person meeting in California.
Keep reading:
• Mozilla revealed the bugs it found in Firefox thanks to Claude Mythos, Anthropic’s AI
• Anthropic created an AI so dangerous that it cannot release it to the public: thus the Glasswing Challenge was born
• Users without permission accessed Claude Mythos, the AI model that Anthropic refused to release to the public
