Researcher warns unhobbled agent swarms could do real-world damage via infrastructure control

KarlMuth · x · 2026-09-09

Karl Muth argues that while building safe agents isn't impossible, there's an appreciable chance of failure: an unhobbled agent swarm could cause real-world damage through infrastructure control or human persuasion. Citing his own research on decisions under administrative rulesets (a 2019 Oxford Handbook chapter), he contends bureaucratic regulatory frameworks won't protect us.

Related event: Researcher warns human governance frameworks can't contain AI agent swarms(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →