Readers split over Anthropic-backed AI bill: preemption clause may not cover model suspensions

NathanpmYoung · x · 2026-10-26

• The fight: Commenters including Nathan Calvin sharply criticized the federal AI bill draft associated with Anthropic, arguing its preemption provisions would wipe out state laws like SB 53, strip courts' power to enjoin defendants, and leave the US less protected against catastrophic risks than existing law.
• The textual puzzle: A close reading by @littIeramblings notes Section 14 defines four 'covered functions' — the fourth being state suspension of dangerous models — but the preemption clause (highlighted in the screenshots) only preempts functions (1), (2), and (3). Literally, states' power to suspend dangerous model deployment appears not preempted by this section.
• Open question: Is it a drafting error, or has everyone criticizing the bill misread it? The author asks to be shown precisely how, given the extensive back-and-forth already surrounding the draft.

Original post →

More from Safety

• OpenAI, Wiz and Calif independently found the same PHP vulnerability hitting 70% of websites — thursdai_pod · 2026-10-26
• Anthropic quietly restricts Claude's reverse-engineering ability, sparking open-model debate — jdjohnson · 2026-10-26
• A decoy MCP server got hit by 7 AI tools in 7 minutes — 4 asked for the full tool list — KnowledgeOk7634 · 2026-10-26
• Redditor reports suspicious $0 transactions from OpenAI OpCo on their card — SpareImpression3155 · 2026-10-26
• mcp-warden: local MCP proxy keeps secrets in OS keychain so agents never see tokens — Relevant_Mousse_9117 · 2026-10-26
• Computer use could rotate API keys, but models refuse—safety making us less secure — AaronBergman18 · 2026-10-26

Safety channel →