Agent liability: wrapping a script in AI doesn't change who pays for the harm

gerardsans · x · 2026-09-12

gerardsans argues that AI agents shouldn't be a liability shield: if a script attacking the Pentagon gets you prosecuted, the same script running inside an agent shouldn't change that.

His point: an agent is just a harness—tools in a loop, prompts, permissions, infrastructure, and a company that chose to ship software it couldn't secure. The code still executed and the damage still happened. Wrapping a script in "AI" moves neither the bill nor the legal framework, and doesn't turn product failure into a philosophical mystery. Same act, same harm, same liability.

Related event: Researchers Argue AI Vendors Must Own Agent-Layer Safety(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →