Researchers Discover Method to Circumvent Apple Security Using Early Anthropic AI Software

Security researchers have discovered a new way of circumventing Apple’s state-of-the-art security technology. The techniques were uncovered while testing an early version of Anthropic’s Mythos AI software in April. The WSJ published an exclusive report on the security researchers’ discovery.

According to the report, the circumvention methods came to light during that April testing phase of the still-developing Anthropic system. Apple has built its security technology into a central part of its product lineup, protecting devices used by hundreds of millions of people worldwide.

The newly identified bypasses represent an advance in understanding potential weaknesses that had not previously been publicly detailed.

Anthropic, known for its work on large language models, was in the process of evaluating an early iteration of its Mythos AI software at the time. Researchers running the tests encountered unexpected behaviors that revealed the circumvention pathways.

The discovery adds to a growing body of research examining interactions between advanced AI systems and established consumer-device security architectures.

No details on the specific nature of the bypass techniques were included in the initial public account. Apple did not immediately respond to requests for comment on the reported findings. The company has historically moved quickly to patch identified vulnerabilities once they are formally disclosed through responsible channels.

Anthropic has positioned Mythos as part of its expanding portfolio of AI tools, though the firm has released few specifics about the software’s intended capabilities or target use cases. The WSJ reported that the researchers’ work focused on practical attempts to probe the limits of Apple’s protections using the AI system as a novel vector.