A cybersecurity startup has publicly demonstrated a method to bypass the security architecture of Apple’s unreleased M5 processor, raising questions about the resilience of modern hardware defenses. The exploit, developed by a California based firm named Calif, reportedly leveraged capabilities of Anthropic Mythos, an advanced artificial intelligence model, to achieve the breach.
The demonstration, confirmed by the startup, marks the first known public instance of an AI system being used to compromise a next generation system on a chip from Apple. While the M5 chip has not yet been officially launched by Apple, industry analysts had widely anticipated its release as a successor to the current M4 series. The breakthrough by Calif suggests that vulnerabilities may exist in the processor’s secure enclave or memory management unit.
According to researchers familiar with the technical details, the approach used by Calif involved feeding the Mythos AI model with architectural specifications and known vulnerability patterns. The model then theoretically identified a potential pathway to circumvent hardware level encryption and memory isolation features. The exploit itself is understood to be a proof of concept, developed in a controlled laboratory environment.
Details of the security breach method
The specific attack vector has not been fully disclosed by the startup. However, industry experts point to the complexity of modern chip security as a key factor. Apple’s M series chips use a custom ARM based architecture with dedicated hardware for secure boot, encrypted memory, and biometric data processing. Mythos AI, developed by a separate startup, is known for its ability to process large codebases and suggest novel exploit chains.
The announcement has generated significant discussion within the cybersecurity community. Security researcher Dr. Emily Zhao, speaking on background, said that while the exploit is likely not replicable on current M4 hardware due to differences in architecture, it serves as a warning for future chip design. “If an AI can map out attack pathways before the hardware is even released, it changes the timeline of vulnerability discovery significantly,” she said.
Implications for consumers and enterprises
For the general consumer, the practical risk remains low at this stage. The exploit was performed on a simulated or early prototype version of the M5 chip, and no public release of the attack code has been made. Apple has not yet commented on the development, and the company typically issues security advisories only after a thorough investigation.
Enterprise users and developers, however, have more cause for concern. If the method can be replicated on future production chips, it could undermine the security of systems handling sensitive data. Apple’s M series processors are used in high end laptops and desktops, often in environments that require robust encryption and isolation.
Industry reaction and future outlook
The disclosure follows a growing trend of AI assisted cybersecurity research. Companies like Google and Microsoft have previously used machine learning to find software bugs. The use of AI to target unannounced hardware is a newer development. Calif has stated that it will present its full findings at an upcoming cybersecurity conference, though the date has not been confirmed.
Apple is expected to release the M5 chip in the next product cycle, potentially in late 2025 or early 2026. Until then, the company has time to review the findings and implement countermeasures in the final chip design. The effectiveness of these countermeasures will be a key focus for security teams and technology investors in the coming months.
Source: GeekWire
