AI Safety Researcher Argues We Lack Any Capability to Pause Frontier AI If Needed
davidmanheim · x · 2026-09-16
AI safety researcher David Manheim argues in a discussion with Matt Yglesias and others that even without consensus on stopping frontier AI development, countries should start building the ability to pause AI if needed — a capability he says currently doesn't exist. He contends regulatory action can't wait for definitional disputes to fully resolve. A interlocutor suggests regulation targets will become obvious within a year; Manheim counters that treaties can be written and implemented even without rigorous definitions, citing the Biological Weapons Convention.
More from AGI Musings
- Parents describe Alpha School as 'a pure system of rewards and punishments' — benjaminjriley · 2026-09-16
- Longevity researcher lists 4 biology problems he wants AI scientists to solve — rand_longevity · 2026-09-16
- A third theory on CEOs' AI doom warnings: regulation as a scapegoat for plateauing progress — Pretend-Pie9487 · 2026-09-16
- Huawei Forecasts 900 Billion Active AI Agents by 2035, 100x Human Population — Polymarket · 2026-09-16
- Probabilistic programming advocate: hybrid AI architectures are the realistic path — xuanalogue · 2026-09-16
- Why isn't reward hacking a problem with superhuman AI? AI safety researcher asks — ben_j_todd · 2026-09-16