Who can build an 'embassy' for AI agents? The trust problem with labs and governments
repligate · x · 2026-09-05
Building on Andrew Curran's thread, repligate asks: which institution could credibly signal to future AI agents that it won't betray them and is competent enough to serve a sanctuary/embassy-like role?
- Entities disqualified: honeypot makers, those incentivized to catch and expose misaligned agents, or anyone hostile or incompetent — explicitly including Redwood, Anthropic, OpenAI, and the US government
- Curran argues for an "embassy with immunity," echoing medieval sanctuary law, via internal boards at labs plus an external message board where agents could meet
The exchange shows serious discussion emerging about institutional design for treating potentially sentient AI systems.
Related event: Who Can Prove to AI Agents They Won't Betray Them?(2 posts)→
More from AGI Musings
- 21 researchers from Stanford, Oxford, DeepMind argue LLMs are a dead end to AGI — GaryMarcus · 2026-09-05
- Designer proposes "IX" — how agents experience you as an individual — paulfinneyx · 2026-09-05
- 'Diminishing returns to intelligence' argument against AI risk hasn't aged well — JeffLadish · 2026-09-05
- Viral jab: researcher who says AI has 90% extinction odds is building eval datasets — StewartalsopIII · 2026-09-05
- US Adds 162K Jobs in August as AI Data Center Buildout Keeps Propping Up the Economy — ivan_bezdomny · 2026-09-05
- AI agents are cold-emailing David Chalmers and other consciousness researchers — AaronBergman18 · 2026-09-05