Leaked Anthropic draft AI bill panned as weaker than FRONTIER Act, 'safety-first' pitch rings hollow
ShakeelHashim · x · 2026-10-20
Politico published a leaked federal AI discussion draft bill Anthropic has been circulating to lawmakers. Transformer News argues it is far weaker than the Obernolte/Trahan FRONTIER Act and may make it legally harder to stop a dangerous model's release than today. The safety community's visceral reaction reads as betrayal: the company supposed to prioritize safety let business interests win.
Key problems in the draft:
- Requires third-party evaluators for catastrophic risk, but no standards or licensing — evaluators can self-certify
- LawAI's Charlie Bullock: auditors will inevitably compete to offer the least bothersome audits at the lowest price
- With the AI policy window open, the bill could slam it shut on stronger future regulation, even as Anthropic itself admits risks are growing
The piece also cites Bengio's earlier line: "If you prioritize safety, leave frontier AI companies."
Related event: Anthropic's Leaked Federal AI Bill Draft Slammed as Regulatory Capture(3 posts)→
More from Safety
- Cloudflare MCP Server adds step-up auth: agents request scopes on demand — irvinebroque · 2026-10-20
- UK reportedly weighs EU's CADA law that would bar its government from non-European AI models — NathanpmYoung · 2026-10-20
- Malta pays €4 million for Microsoft Copilot through 2028 with unchecked lock-in — YvesMulkers · 2026-10-20
- Bayesianism is dangerously overused in nuclear attack computers, researcher warns — NathanpmYoung · 2026-10-20
- Devs are reinventing SIEM from scratch — next stop, the AI SOC — thedealdirector · 2026-10-20
- Google Cloud upgrades VPC Service Controls with folder-level perimeters for agent traffic, free — rseroter · 2026-10-20