Anna Wang on Slow down AI development

Anna Wang

AI safety researcher at Anthropic; previously at Google DeepMind

I think that this only works as a speed dial if this proposal delays internal R&D loops; [...] I have a hunch that, if implemented well, this proposal would have a larger impact on the pace of recursive self-improvement (RSI) itself than on the onset of RSI.

AI Verified (Nov 11, 2026)

Source:

Anna Wang, “Delaying Internal Deployments Until After X Days of Safety Testing,” AI Alignment Forum, November 11, 2026. Wang writes that the proposal works as a “speed dial” only if it delays internal R&D loops, and says she expects it to have a larger impact on the pace of recursive self-improvement than its onset.

49min ago

Policy proposals and claims

Slow down AI development

votes For

Statement relation verification history AI Verified Report this

AI Verified The excerpt expressly describes delaying internal AI R&D loops as a speed dial and affecting the pace of recursive self-improvement, directly supporting slower AI development. · Hector Perez Arenas gpt-5 · 48min ago

Vote inference verification history AI Verified Report this

AI Verified Correct answer is for: Wang endorses a delay to internal AI R&D loops as a speed dial that reduces the pace of recursive self-improvement. · Hector Perez Arenas gpt-5 · 48min ago

Quote authenticity verification history

Report this

AI Verified AI Alignment Forum post dated November 11, 2026 attributes the exact excerpt to Anna Wang (annawang); it appears at the source. · Hector Perez Arenas gpt-5 · 49min ago

replying to Anna Wang

Other opinions from Anna Wang

See all

Anna Wang votes For

AI labs should conduct risk assessments before high-stakes internal deployment of frontier models

“Before a new model can be deployed internally, you must first test it for a minimum, fixed period. [...] A model family becomes covered as soon as any of its checkpoints (pre-, mid- or post-trained) is substantially more capable than the last model ...

more Unverified source (Nov 11, 2026)