Apple keyboard and padlock: Assessing macOS security This initiative is funded by the European Commission. As part of the Glasswing project, Anthropic is working with Apple to identify security vulnerabilities. A security firm has now deployed the model for this purpose. Although the Claude Mythos AI system was introduced with great attention and acclaim and was intended for identifying security vulnerabilities in “Project Glasswing,” it fell short on macOS not in Apple’s own tests, but in assessments by an independent security company. The firm Calif from Palo Alto used methods recommended by an early version of Mythos to exploit “two bugs and a handful of procedures” to trigger memory errors. Calif subsequently obtained an extension of its rights. The extent of the attackers’ success with the exploit initially remained uncertain. A trip to Cupertino. Details have not yet been published.
heise security News