AI safety researcher: frontier training needs government shutdown capability, current models already unmanaged
davidmanheim · x · 2026-11-19
In a debate over pausing frontier AI training, safety researcher David Manheim restates his long-held position: he calls for shutting things down precisely because any shutdown regime will take years to implement, so governments need the ability to monitor frontier training runs and intervene if necessary. He argues that halting further training wouldn't address risks from models already deployed — loss-of-control, inadequate human oversight, and cybersecurity risks — all of which we are 'obviously not currently on track to manage safely.'
More from AGI Musings
- Tim Draper says AI will soon eliminate late-stage VCs who are 'just investment bankers' — arian_ghashghai · 2026-11-19
- AI safety researcher: governments must be able to monitor and halt frontier training — davidmanheim · 2026-11-19
- UK tech figure slams defeatism on homegrown frontier models: 'We must be mad not to try' — rosstaylor90 · 2026-11-19
- Keynes' 1930 warning on technological unemployment is landing right on schedule — tobowers · 2026-11-19
- Researcher defends Anthropic: any careful game-out of the dynamics points its way — AaronBergman18 · 2026-11-19
- Erik Hoel: consciousness is a great mystery, but its definition isn't — experts largely agree — koltregaskes · 2026-11-19