Two leading AI labs are lobbying for opposite versions of the same state's safety framework
OpenAI has announced its backing of a revised AI safety bill in Massachusetts, one narrower in scope than an alternative framework backed by Anthropic. The divergence reflects different theories about how AI companies should be held accountable as state legislatures move faster than Congress.
What the bills require
The version OpenAI backs centers on transparency and watermarking requirements, along with third-party audits of AI systems. The alternative, which Anthropic has backed, would mandate independent safety evaluations, a mechanism that places more distance between a company and the entity assessing its risks.
The OpenAI-backed bill does not impose strict liability on foundational model developers. Liability provisions are typically the teeth of any regulatory framework. Without them, disclosure requirements and audits function more as reputational constraints than legal ones.
A corporate split with deliberate logic
Get the latest model rankings, product launches, and evaluation insights delivered to your inbox.
OpenAI, which has faced sustained scrutiny over its governance structure and commercial trajectory, has an interest in shaping regulation that it can absorb without disrupting its deployment pace. Anthropic, which has built its public identity around safety-first engineering, has an incentive to back frameworks that validate the kind of rigorous evaluation it already conducts internally.
Companies that invest heavily in internal safety infrastructure have a rational preference for regulations that require competitors to meet similar standards. Companies that have not made those investments prefer frameworks that leave more discretion to the developer.
What state-level action can do
Massachusetts joining the patchwork of state AI legislation creates compliance headaches for companies operating nationally, but it also creates a testing ground. California's earlier AI regulatory battles showed that state bills can shift industry norms even when they fail to pass. The lobbying process itself forces companies to articulate positions they might otherwise leave vague.
The bill OpenAI backs would require AI companies to publish safety frameworks. If passed, that requirement would generate a public record against which future incidents could be measured. Whether that record would carry legal weight depends on provisions the underlying legislation has not yet detailed.
The Massachusetts legislature has not yet voted on either version. That vote will determine whether the OpenAI-backed framework advances, whether Anthropic's preferred approach gains traction, or whether the two companies produce a compromise that neither fully endorses.