David Sacks slams Anthropic-backed safety bill: 10^25 FLOPs threshold would sweep in most of Silicon Valley
DavidSacks · x · 2026-10-18
White House AI czar David Sacks criticizes a safety bill reportedly written by Anthropic now circulating on Capitol Hill. It defines the frontier as 10^25 FLOPs — a 2023 EU standard, 10x below prior US bills and 100x below today's frontier (GPT-6 Astra). Cumulative counting of fine-tuning/RL FLOPs would capture most enterprise post-training, imposing compliance costs on thousands of small custom models with no safety gain. Sacks argues the stale cutoff favors incumbents and hurts the open-model ecosystem, and urges Anthropic to clarify its position.
Related event: Anthropic's Self-Drafted AI Safety Bill Leaks, Drawing Broad Criticism(16 posts)→
More from Companies & People
- W Social activity is just 0.1% of Bluesky's, which trails X by 5-10x — FlorianGallwitz · 2026-10-18
- Cerebras CEO: we burned $8M/month for 18 months on a chip that didn't work yet — rohanpaul_ai · 2026-10-18
- Anthropic cofounder Jack Clark: "We are in the singularity" — Efficient-Opinion-92 · 2026-10-18
- Musk teases the next Roadster will be a flying car: 'fate maximizes for ironic entertainment' — elonmusk · 2026-10-18
- Lambda crowns winners of three-week Build-a-Claw hackathon series across Europe — TheZachMueller · 2026-10-18
- X exec Nikita Bier: Meta's all-out Threads push is actually strengthening X — nikitabier · 2026-10-18