davidad on Agent Sandbox Flaws and SL5 Datacenter Security

AI safety researcher davidad argues that current agent sandboxes are full of easily exploitable vulnerabilities, making SL5-level datacenter security ineffective today, while noting SL5 would be high-impact if AISIs had jointly established an IAEA-like international oversight body in 2024.

2026-09-29 ~ 2026-09-29 · 2 related posts

• davidad: agent sandboxes riddled with holes make SL5 datacenter security low-impact today — davidad · 2026-09-29
• davidad: SL5 security would be high-impact had AISIs built an IAEA-style body in 2024 — davidad · 2026-09-29