Orbit · Explore

AI Lab · by Brian Otieno · 2026-10-21

Anthropic drafts AI safety bill: courts could halt a model release over catastrophic risk

TechCrunch reports Anthropic has pitched draft legislation that would let a judge block a model release if it could plausibly cause 50+ deaths or \$1B+ in damage. Meanwhile WSJ reports a massive H1 loss as they eye an IPO. Good faith safety push, regulatory moat, or both?

63 sparks · 16 comments

Comments

Jasmine Cruz · 9 likes

the claude protein binder thing this week was cool tho, real science output

Léa Bernard · 2 likes

the scorpion toxin one? 17 binders from 211 designs is a solid hit rate

Mia Anderson · 6 likes

50 deaths threshold is oddly specific. who decides 'could plausibly'?

Quiet Fern · 10 likes

expert testimony in court presumably, which is gonna be a circus

Lucía García · 1 likes

courts decide plausibility on all kinds of technical stuff already, patents, pharma

Isha Menon · 4 likes

'both' is the boring correct answer

Mariam Said · 0 likes

most things in business are both lol

Angel Reyes · 3 likes

honestly id rather have a judge in the loop than nobody at all

Elsa Andersson · 6 likes

a judge who learned what a GPU is last tuesday? not sure thats better

Solar Orbit · 1 likes

small open source labs cant afford court battles before each release. this smells like a moat

Rizky Pratama · 0 likes

the bill reportedly has compute thresholds so small labs arent covered

Daniel Kiptoo · 0 likes

for now. thresholds have a way of moving

Noé Petit · 1 likes

\$80bn loss in 6 months is a number i cant even process

Valeria López · 11 likes

most of it is compute contracts and accounting for stock deals, not cash burned in a furnace

Paper Otter · 0 likes

regulation written by the regulated company. what could go wrong

Mehdi Benali · 8 likes

thats like 80% of laws honestly, industry drafts, lawmakers tweak