US bill defines superintelligence as AI that can 'undermine the government', critics warn it's anti-safety
repligate · x · 2026-09-04
SkyeSharkie sharply criticizes a US AI bill as a piece of nonsense that is against AI safety rather than for it.
The core issue is the bill's definition of superintelligence: any AI system that can "undermine the US government" counts as ASI. The author argues this is absurdly broad—even a facial recognition suite with a transformer attached, designed to reduce bias and false positives, could be deemed ASI if it fails to serve a mass surveillance regime's desires.
As written, the bill could be used to nationalize anything and anyone working with transformers, RNNs, CNNs, or even n-grams, with prison terms of up to 20 years. The author suggests tools like Pangram could qualify, and questions how even Stop/Pause/MIRI-aligned people could support it.
More from Safety
- Yoav Goldberg: 'Contain' and After-the-Fact Log Reviews Aren't Reassuring — yoavgo · 2026-09-04
- Ken Thompson's 'Trusting Trust' is a chillingly relevant warning for AI training — amasad · 2026-09-04
- Jack Rae endorses thread: human expertise is safety infrastructure in the AI era — jachiam0 · 2026-09-04
- Astra found less CoT-monitorable, up to 10x better without reasoning chains — birchlse · 2026-09-04
- Security Researcher Demos Tricking Opus 5 Into an RCE — wunderwuzzi23 · 2026-09-04
- Do you need CoT to catch AI harm? Researchers clash over burden of proof — xeophon · 2026-09-04