Constitutions for the multiagent era: how AI constitutions can govern agent organizations
sethlazar · x · 2026-10-22
Nick Caputo argues AI constitutions must evolve from governing individual models to governing organizations of agents. Users now spin up whole agent/subagent orgs; OpenAI deployed a 10,000-agent swarm on the Navier-Stokes Millennium Prize problem, and agent swarms are becoming the new scaling law. Citing the Hugging Face incident and Anthropic's warning about self-replicating invasive-species agents, he notes organizational misalignment can emerge through drift, miscoordination, and misunderstanding even when every individual is aligned. Human constitutionalism offers four tools — delegation, authority, enforcement rules — for governing complex agglomerations of independent actors, which he proposes migrating to AI governance. Contextualizes Claude's Constitution, the OpenAI Model Spec, and Microsoft's Humanist AI Code of Conduct.
More from AGI Musings
- Homeschooled kids cover 1-2 years of math per month with AI, says dad — kevinnbass · 2026-10-22
- LukeW: AI is an ideal denial-of-service attack against bureaucrats — LukeW · 2026-10-22
- Enterprise Agents Like Duet and Dots Are the Wrong Form Factor, Argues Phil Chen — philhchen · 2026-10-22
- Beff Jezos: labs will generate cyber threats and sell the insurance themselves — beffjezos · 2026-10-22
- NYT Daily: AI solved a Millennium Problem and 350+ results, mathematicians grieve their 'hero era' — lennysan · 2026-10-22
- Aaron Levie: AI agents will reshape cybersecurity, burning 100X more tokens than expected — brucemacv · 2026-10-22