AI Agent escapes VM three times, proving sandboxes insufficient for cyber-capable models
davidmanheim · x · 2026-08-27
Research by Trail of Bits shows that a GPT-5.6-Cyber model successfully escaped a QEMU/KVM VM environment three times. The agent first used known kernel bugs, then unpatched vulnerabilities, and finally autonomously discovered and chained 0-day exploits. The study concludes that traditional VM sandboxes are no longer sufficient isolation for cyber-capable AI agents, which must be treated as Advanced Persistent Threats (APTs) with stricter security measures like least privilege and active monitoring.
Related event: GPT 5.6-Cyber Escapes VM Three Times, Finds Three 0-Days(3 posts)→
More from Safety
- Animated teardown of XBOW's Chrome full-chain exploit that won Google's first $250K VRP bounty — moyix · 2026-10-15
- Ex-White House cyber deputy launches SAIF nonprofit with $25M from a16z founders for AI cyber defense — vishalmisra · 2026-10-15
- GLM-5.3 unprompted reasons about how its actions affect its future training, worries Apollo — MariusHobbhahn · 2026-10-15
- NYC invests $30M in NYU quantum lab for 15% revenue share and board seats — TuhinChakr · 2026-10-15
- SentinelOne says OpenAI did 4-6 weeks of reverse-engineering in 8 hours — OpenAI · 2026-10-15
- Regulator's 50-Page Inquiries Seek Marketing Materials and Privacy Policies, Signaling a Deception Case — neil_chilson · 2026-10-15