Palo Alto: A cybersecurity startup has claimed a major breakthrough that could reshape concerns around AI-driven cyber threats, stating it used an advanced artificial intelligence model to bypass security protections in Apple’s latest M5 chips within days.
The development, reported by Palo Alto-based startup Calif, involved leveraging a preview version of Anthropic’s Claude Mythos model to create a working exploit targeting Apple’s newly introduced Memory Integrity Enforcement (MIE) system.
Exploit challenges Apple’s latest security architecture
Apple has long been known for its tightly controlled hardware-software ecosystem, which has helped it build a strong reputation for device security. The introduction of Memory Integrity Enforcement (MIE) was seen as a significant leap forward in hardware-assisted protection against memory corruption attacks.
MIE is designed to detect and block attempts to manipulate system memory at the chip level before they can be exploited. According to industry estimates, Apple spent nearly five years and potentially billions of rupees developing this security layer.
However, Calif researchers claim they successfully bypassed these protections by combining two known bugs with multiple exploitation techniques. The result, they said, was a macOS kernel memory corruption exploit capable of surviving MIE safeguards—something not previously demonstrated publicly.
Exploit reportedly built in under a week
One of the most striking aspects of the claim is the speed at which the exploit was developed. According to Calif, the process unfolded rapidly over just a few days.
Researcher Bruce Dang reportedly discovered the vulnerabilities on April 25. Within days, experts including Dion Blazakis and Josh Maine collaborated to build the exploit, achieving a working version by May 1.
The company described the discovery as partly accidental but emphasised that the AI model significantly accelerated the process by identifying patterns in known vulnerability classes.
Role of AI in accelerating cyber exploits
Calif credited Anthropic’s Mythos Preview model as a key enabler in the exploit’s development. According to researchers, the AI system demonstrated an ability to generalise across known exploit categories, rapidly identifying weaknesses once it recognised familiar patterns.
This highlights a growing trend in cybersecurity where AI tools are not only used for defence but are increasingly capable of assisting offensive research.
However, the company clarified that human expertise remained critical, particularly in bypassing Apple’s newly designed MIE protections, which required nuanced understanding beyond the AI’s capabilities.
Concerns over AI-human collaboration in hacking
The findings raise broader concerns about the implications of combining advanced AI systems with skilled cybersecurity professionals. Calif said the project was designed to explore exactly this intersection.
The ability to produce a sophisticated kernel-level exploit in less than a week suggests that AI could dramatically reduce the time and effort traditionally required for such attacks.
This could potentially lower the barrier for advanced cyberattacks, especially if such tools become more widely accessible.
Responsible disclosure and industry response
Calif stated that it shared its findings directly with Apple through an in-person meeting at the company’s California headquarters, rather than submitting the report through conventional channels.
Meanwhile, Anthropic has taken a cautious approach to the release of its Mythos model. After internal testing revealed its ability to autonomously identify and exploit vulnerabilities, the company restricted access to a limited group of organisations under its Project Glasswing initiative.
The model’s capabilities have already drawn attention across the industry. During testing, it reportedly identified hundreds of vulnerabilities in Mozilla’s Firefox browser, further underscoring the rapid evolution of AI in cybersecurity.
Conclusion
The reported exploit against Apple’s M5 security protections marks a significant moment in the intersection of artificial intelligence and cybersecurity. While Apple’s defences remain among the most advanced in the industry, the incident highlights how quickly AI-powered tools are changing the threat landscape.
As companies continue to develop stronger protections, the parallel advancement of AI-driven exploit techniques may create an ongoing arms race between security systems and those seeking to bypass them. Ensuring responsible use of such powerful AI models will be critical in maintaining digital security in the years ahead.