XBow details AI exploit reliability: ~10% raw page-flip rate, 48 retries per attempt
moyix · x · 2026-10-15
moyix shared technical details behind XBow's AI-agent crashpad exploit: with retries built in, they don't have per-attempt success numbers; the raw page-flip transition succeeded only 10% of the time (10,175/100,000), but 48 tries per attempt made the exploit quite reliable. Restricting the goal to a single write gave the agent a clear objective that was easier to satisfy, and the existing WCP table base technique was sufficient to reach arbitrary code execution.
Related event: moyix Details How XBow's AI Agent Autonomously Developed V8 Sandbox Escape(7 posts)→
More from coding & agent
- Omarchy bug: missing rtkit causes Chrome/Electron apps to get SIGKILLed — fix inside — mark_k · 2026-10-15
- Google's DuoDiscover tops NVIDIA SOL-ExecBench, 2.68x faster than production TPU kernels — lateinteraction · 2026-10-15
- AIPOCH open-sources a local AI research workbench with 548 science connectors and reproducible .science packages — FellMentKE · 2026-10-15
- Agent stack check: harnesses, OTel telemetry and MCP are settled — evals remain a mess — _ScottCondron · 2026-10-15
- Claude Code's safety guardrails refuse even non-sensitive env var copies, user complains — Promptmethus · 2026-10-15
- Fragment raises $12M Series A to make AI agents safe financial actors — mattturck · 2026-10-15