Joshua Saxe's Keynote: Layered Defenses for Securing AI Agents
joshua_saxe · x · 2026-08-17
Joshua Saxe shares his keynote on securing AI agents that make dangerous decisions autonomously, arguing that alignment, sandboxing, and monitoring are each necessary but insufficient, and together can satisfice safety if autonomy is increased slowly and controllably.
Related event: Joshua Saxe: Layered Defenses Needed to Keep Autonomous AI Agents Safe(2 posts)→
More from AGI Musings
- Hugo Award Winner: Benevolent AI Trades Help for Cat Pictures — DynamicWebPaige · 2026-08-17
- AI safety experts urge shift from personality-based trust to institutional regulation — Miles_Brundage · 2026-08-17
- AI chips will replace crypto as the superior energy-to-value transducer — beffjezos · 2026-08-17
- Opinion: Open source AI should focus on improvement over hype — bindureddy · 2026-08-17
- Humans are the moat: Why AI cannot replace real interaction — xiaosun86 · 2026-08-17
- User laments Claude's shift,质疑 Anthropic's strategic direction — markjeffrey · 2026-08-17