2 hrs ago
Anthropic Backs Massachusetts AI Safety Bill, OpenAI and Google Oppose
Massachusetts lawmakers are considering a rule about checking powerful AI systems for dangerous problems.
The proposal would require large AI companies to use independent reviewers every four months.
Anthropic supports the proposal and says frequent checks are needed because AI models change several times a year.
OpenAI and Google oppose the bill.
OpenAI prefers yearly outside audits instead.
OpenAI also worries that different rules in each state would make compliance expensive and slow down useful tools.
Anthropic’s support may also help its business because it already emphasizes safety.
The decision could affect how other states regulate AI.
A Massachusetts Senate proposal would require large AI developers to commission independent catastrophic-risk evaluations every four months.
Anthropic supports the measure and favors independent risk assessments every four to six months.
OpenAI and Google oppose the bill, while OpenAI favors annual third-party audits similar to Illinois’s approach.
OpenAI argues that differing state regulations could create costly compliance burdens and delay beneficial AI tools, including cybersecurity products.
The dispute could influence future state AI laws because the United States currently lacks a comprehensive federal framework.
- Who
- Anthropic supports the Massachusetts proposal; OpenAI and Google oppose it, while Massachusetts senators are considering it.
- What
- A Senate proposal would require large AI developers to commission independent evaluations of catastrophic risks in their models every four months.
- Where
- Massachusetts, United States.
- When
- The articles do not provide a specific date; the proposal is under consideration.
- Why
- Supporters want evaluations to keep pace with rapidly changing AI models, while opponents cite concerns about inconsistent state regulations, compliance costs, and delays to useful tools.
Anthropic and Frequent Evaluations
OpenAI and State-Regulation Concerns
How often models should be assessed
Anthropic and Frequent Evaluations
Anthropic supports independent risk assessments every four to six months, matching the release cycle of frequently revised frontier models.
OpenAI and State-Regulation Concerns
OpenAI favors annual third-party audits, similar to the approach taken in Illinois, and opposes the Massachusetts bill.
State-by-state regulation
Anthropic and Frequent Evaluations
Frequent state-level requirements can provide binding oversight while no sufficiently comprehensive federal framework exists.
OpenAI and State-Regulation Concerns
OpenAI argues that conflicting rules across states would create significant compliance costs and could delay beneficial models, including cybersecurity tools.
Commercial incentives
Anthropic and Frequent Evaluations
Anthropic says rigorous evaluations are important for AI safety, though the requirement may also reinforce its safety-focused market position.
OpenAI and State-Regulation Concerns
Opponents’ concerns focus on regulatory burden and fragmentation; the article does not state that OpenAI or Google oppose third-party review itself.
Key facts
- Proposal
- A Massachusetts Senate measure included in a larger economic development bill.
- Evaluation frequency
- Independent catastrophic-risk assessments every four months under the proposal.
- Anthropic’s position
- Supports the bill and favors independent assessments every four to six months.
- OpenAI’s position
- Opposes the bill and favors annual third-party audits.
- Google’s position
- Opposes the Massachusetts proposal.
- Central dispute
- The cadence and independence of third-party safety evaluations, not whether outside review should exist.
- Broader significance
- The Massachusetts approach could be copied or adapted by other state legislatures.
Sources
Related news
House Democrats Press Anthropic, OpenAI Over Rogue AI Agents
Anthropic CEO Dario Amodei rejects claim AI regulation concentrates power
AI Labs Score Poorly on Rogue-Model Control Measures
Judge Blocks Pentagon’s Anthropic Supply Chain Risk Designation
Anthropic hires former California Supreme Court justice amid AI incidents
Anthropic CEO Defends AI Warnings Amid Trust Crisis
Eight Shot, One Killed at Taunton House Party
Anthony Bourdain’s Food Lessons Reach Beyond Tourist Destinations
What's Next for Lindsay Clancy After Her Mistrial?
Lindsay Clancy Mistrial Leaves Three-Child Murder Case Unresolved
