AI BriefingClaude Mythos Takes 5 Days to Develop a macOS Exploit
Key point
Anthropic's Claude Mythos helped develop a macOS exploit in 5 days, and Apple has fixed the flaw.
Details
The unreleased frontier model Claude Mythos Preview was used in real-world exploit development during macOS security research.
Security firm Calif used Mythos to build a kernel privilege escalation chain linking two vulnerabilities on Apple M5 with Memory Integrity Enforcement (MIE) enabled, running macOS 26.4.1. According to the company, the bug was discovered on April 25, a working exploit was completed by May 1, and the entire effort took about 5 days.
- Calif stated that Mythos accelerated vulnerability identification and exploit development, but bypassing MIE still required design work from human researchers.
- Apple fixed CVE-2026-28952 in the macOS Tahoe 26.5 security documentation, crediting the discovery to Calif.io in collaboration with Claude and Anthropic Research.
- Anthropic is distributing this model in a limited manner under Project Glasswing, and is working with Apple, Google, Microsoft, and others on defensive cybersecurity applications.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
US Appeals Court Upholds Pentagon Blacklisting of Anthropic as Supply Chain Risk
Analysis of the 'Inference Gap' Between Claude's Reduced Reasoning Tokens and Benchmarks