The two companies that set the pace of the AI industry are publicly fighting each other over how their industry should be regulated, and the fight is happening in a state legislature.
Anthropic and OpenAI have taken opposing positions on a proposed AI safety law in Massachusetts, according to Bloomberg. The state is pushing one of the strictest AI regulatory bills in the country, and the two leading labs disagree sharply over how much regulation is appropriate and where liability should fall. Their public lobbying has become a battle over the shape of the law, with each side pressing legislators to adopt rules that fit its business model.
The dispute breaks an industry consensus that had held for years. AI companies have generally argued for some form of regulation, presenting themselves as responsible stewards of a powerful technology. That unity was always thinner than it appeared. Now, with a concrete bill on the table, the differences have become explicit, and the industry’s public position has fractured along business lines.
The Massachusetts bill is among the most ambitious state efforts to regulate AI. It would impose requirements on the developers of large models, covering testing, disclosure and liability for harms that models might cause. Supporters say the rules are needed to protect consumers and society from risks that the federal government has failed to address. Opponents say the requirements are unworkable, would slow innovation and would push AI development to states with friendlier rules.
Anthropic’s position reflects its long-standing advocacy for oversight. The company has built its public identity around safety, publishing frameworks for responsible scaling and arguing that frontier labs should accept binding commitments. In Massachusetts, it has pushed for rules that it says would establish clear standards without strangling development. Its approach to regulation has been to embrace it early, in the hope of shaping rules that it can live with.
OpenAI’s position is more skeptical. The company has grown into the largest and most commercially aggressive of the labs, and its leadership has argued that premature or poorly drafted regulation could cripple American AI leadership. In Massachusetts, it has objected to provisions that would assign liability to developers for uses of their models, arguing that responsibility should sit with the companies that deploy the systems in specific contexts. The difference is not abstract; it determines who pays when a model causes harm.
The commercial logic behind the split is clear. Anthropic’s business depends on enterprise trust, and it has marketed itself as the safe choice for businesses worried about liability. Strong regulation, with clear rules, helps it sell that story. OpenAI, with the broadest product line and the most at stake in rapid deployment, has more to lose from restrictions and liability rules that could slow adoption or raise its costs. Each company is advocating for the regulatory environment that favors its position.
The fight has real consequences for the bill’s prospects. Legislators considering a complex technology law face dueling testimony from the two most credible voices in the industry, each claiming to represent the responsible position. Lobbying from both sides has intensified as the session has progressed, and the outcome will depend on which arguments the state’s lawmakers find more persuasive. Other states are watching; Massachusetts could set a template for AI regulation across the country.
The dispute also exposes the limits of self-regulation. If the industry’s leading companies cannot agree among themselves on basic questions of liability and oversight, the argument that the industry can police itself becomes harder to sustain. Regulators in Washington and other states may conclude that the safest path is to write their own rules, informed by the industry’s failure to reach consensus.
For now, the Massachusetts bill remains in play, and the two labs remain publicly opposed. The debate has moved beyond technical questions of model evaluation into the core economics of the AI business: who is responsible when AI goes wrong, and who pays. The answers, if the bill passes, will set a precedent that the entire industry will have to live with.
The lobbying has spilled into public view in unusual ways. Both companies have run advertising and outreach campaigns in the state, published position papers and dispatched executives to testify before legislative committees. Their differences have also surfaced in technical debates that outsiders rarely see, such as how model evaluations should be conducted, who should have access to the results and what thresholds should trigger additional scrutiny. Those details, seemingly arcane, determine whether the law has any practical effect. A requirement that only the largest models be tested would bind both labs; a requirement that all models be tested would bind the entire industry. The two companies’ positions on such questions track their sizes and strategies, and legislators have been forced to adjudicate between two well-funded, well-credentialed adversaries who agree on almost nothing except that they should set the rules rather than have them set for them.
