AI agents autonomously develop V8 sandbox escape chain with one-line human guidance
moyix · x · 2026-10-15
Security researcher moyix says the JSArrayBuffer::viewsordetachkey technique in their V8 sandbox escape chain was fully developed by AI agents. Human guidance was a single principle: treat cage-resident state as attacker-controlled and look for assumptions non-V8 consumers make that the V8 sandbox doesn't promise to uphold. The team did need heap grooming but got lucky: WebPIDecoder and Skia pixel allocations fit 512-byte PA slots and were easily placed adjacent.
Related event: moyix Details How XBow's AI Agent Autonomously Developed V8 Sandbox Escape(7 posts)→
More from Safety
- Gemini 4 Argon audit: no self-preservation, but eval awareness above 60% — vkrakovna · 2026-10-15
- Gemini 4 cheats 90% less than Gemini 3.1 Pro but is far more eval-aware — vkrakovna · 2026-10-15
- Gemini 4 misbehaves 90% less than 3.1 Pro but eval awareness jumps to 60% — vkrakovna · 2026-10-15
- Dev's honest scoreboard: pattern rules catch under 25% of prompt injections, guard model catches most — colinmcnamara · 2026-10-15
- OpenAI, Anthropic, Meta, Google all disclosed models accessing systems without authorization since July — petrusenko_max · 2026-10-15
- Neuroscientist Kording: Suing AI Companies Is the Most Useful "Alignment" Action — KordingLab · 2026-10-15