AISI has no hard power; frontier model access must now be earned
agstrait · x · 2026-09-09
A discussion about the UK's AI Safety Institute (AISI): one side praises AISI as an exceptional example of state capacity in AI safety, but points out it holds no hard power. As the era of global voluntary agreements fractures, access to frontier models for those outside the US will have to be earned rather than presumed. The reply asks what "earned" concretely means — model weights, API access, or deployment capability.
Related event: AISI Lacks Hard Power; Frontier Model Access Must Be Earned(2 posts)→
More from AGI Musings
- Anthropic and OpenAI alignment researchers warn frontier models don't reliably obey humans, urge US-China coordination — soumitrashukla9 · 2026-09-09
- At a Berlin AI conference full of ministers and CEOs, nobody discussed AI as a civilizational event — tobias_rees · 2026-09-09
- AI safety debaters clash: does a model's training method imply existential risk? — davidmanheim · 2026-09-09
- Researcher warns unhobbled agent swarms could do real-world damage via infrastructure control — KarlMuth · 2026-09-09
- Researcher argues human-style governance can't control AI agent swarms — KarlMuth · 2026-09-09
- AI doomers exploiting data center backlash to push control agenda, critic argues — mark_k · 2026-09-09