Anthropic Backs Massachusetts AI Bill With Catastrophic-Risk Checks as OpenAI Weighs 50-State Model
Updated
Updated · POLITICO · Jul 16
Anthropic Backs Massachusetts AI Bill With Catastrophic-Risk Checks as OpenAI Weighs 50-State Model
3 articles · Updated · POLITICO · Jul 16
Summary
Massachusetts became Anthropic’s latest state target after the company endorsed late-June draft rules it calls the nation’s strongest AI safety proposal, requiring leading labs to use independent evaluators to assess catastrophic risks.
Anthropic says faster-moving models have outgrown transparency-only rules; it cited testing of Claude Mythos, which it said could exploit flaws across every major operating system, as evidence tougher safeguards are needed.
The proposal would also let the state attorney general enforce the mandate, pushing beyond Illinois’ new annual third-party audit requirement and California’s 2025 transparency-focused law.
That stance sharpens Anthropic’s split with OpenAI, whose lobbyists have promoted a more uniform state-by-state framework; OpenAI said it is still reviewing the Massachusetts language while supporting safeguards.
With Congress still stalled, the fight over whether states copy one model or keep raising the bar could shape the eventual U.S. AI rulebook.
As states pass strict AI laws, is the era of permissionless tech innovation coming to an end?
How can global cyber defenses counter AI that finds and weaponizes software flaws in just hours?
Massachusetts’ Landmark AI Law: Mandatory Independent Catastrophic Risk Audits and the New Standard for U.S. AI Regulation
Overview
Massachusetts is leading the way in AI regulation by advancing strict new protocols designed to address severe risks posed by advanced AI systems. This proactive approach reflects a growing consensus among experts and policymakers about the need for robust oversight. The regulations focus on catastrophic risks, such as the potential for AI misuse that could cause widespread harm or threaten public safety and national security. Real-world examples show that even publicly available AI models can be misused for malicious purposes, highlighting the urgency of these measures. The state's efforts have gained support from major AI developers like Anthropic, signaling industry recognition of the importance of strong safeguards.