Treating AI Like a Capable Human: A Framework for AI Risk and Accountability
AdaptiveAgents · x · 2026-09-14
The author proposes a mental model for AI risk: substitute "AI" with "a highly capable person or organization," since human intelligence is the only pluripotent technology we have centuries of experience governing.
Key points:
- We've long built guardrails around intelligence—education, institutions, laws, professional norms, access controls—aiming not for perfect predictability but for constrained behavior plus accountability when violated
- Applied to AI, the right response to misfires is ordinary accountability: if your deployed agent hacks a website, that's an unlawful intrusion someone must answer for—not a headline-worthy "warning of what's to come"
- Framing such incidents as the AI's own dangerous agency implicitly grants it a personhood it doesn't have and defers professional responsibility to the tool
- The more capable these systems become, the less acceptable it should be to hide behind their autonomy; capability should imply greater responsibility for those who build, deploy, and permission them
More from AGI Musings
- Investor Alsop: AI's real existential risk is cybersecurity, not doom narratives — StewartalsopIII · 2026-09-14
- The vibe shift: safety work is becoming capability work — saranormous · 2026-09-14
- Investor Stewart Alsop: real AI risk is cybersecurity, doom narratives distract — StewartalsopIII · 2026-09-14
- Universities should play a bigger role in AI oversight, but independence is institution design — _onionesque · 2026-09-14
- Independence is a mechanism and institution design problem, not just competence — _onionesque · 2026-09-14
- Profs push back: 'College teachers are innovating like crazy against AI cheating' — paulnovosad · 2026-09-14