Anthropic's Mythos AI discovers a macOS kernel exploit in five days
What Happened
Security blogger Bruce Schneier wrote up a case where an independent researcher used Anthropic's Mythos system to discover a kernel memory-corruption vulnerability on Apple's M5 silicon, end-to-end, in about five days. Schneier connects it to a broader essay arguing that advanced AI is now meaningfully shifting the offense/defense balance in cybersecurity and regulatory systems.
My Take
Every CISO budget conversation for the next year just got harder. If a single researcher plus an API key can produce a kernel exploit in a workweek, the assumption that "we have time to patch" is dead — and so is the assumption that nation-state-grade capability stays scarce. The corporate response will be ugly and expensive: real-time patching pipelines, AI-on-AI defensive monitoring, and a lot of vendors suddenly claiming to do both.
Read Original Source