Pennsylvania lawmakers weigh two bills to regulate frontier AI risks

Pennsylvania’s House Communications and Technology Committee heard initial presentations Monday on two bills aimed at the largest AI developers, as lawmakers argued that states are moving to set rules while federal action remains limited. The measures are proposals, not enacted requirements.

Rep. Melissa Shusterman’s HB 2800 and Rep. Manuel Guzman’s HB 2829 would apply to companies with more than $500 million in annual gross revenue developing models trained with more than 10²⁶ operations, according to a bill analysis by the Transparency Coalition. Both would require covered companies to establish and publish safety frameworks, arrange annual independent audits and report critical incidents to state authorities within 72 hours of forming a reasonable belief one occurred. Imminent risks of death or serious injury would trigger notification within 24 hours.

The bills also call for public disclosures about model releases and protections for employees who report safety concerns. HB 2829’s notable addition is a prohibition on deployment when a developer’s own assessment finds an unreasonable risk of catastrophic-scale harm. The bills define catastrophic risk as more than 50 deaths or injuries, or more than $1 billion in damage, the coalition says.

The proposals differ in other respects, too. HB 2800 would be enforced by the Pennsylvania Emergency Management Agency in consultation with the attorney general; HB 2829 assigns enforcement to the attorney general and preempts local regulation of frontier models. Both provide for penalties and do not create a private right of action, according to the coalition’s comparison.

At the hearing, Shusterman said AI development is advancing rapidly and argued that products capable of serious harm should be subject to standards and government oversight. Guzman framed incident reporting as a basic accountability measure, saying the state should not have to rely on companies to decide whether to disclose a system failure. Their remarks reflect the sponsors’ case for the bills, not findings that the cited scenarios have occurred in Pennsylvania.

Steve Wimmer, a policy and technical adviser with the Transparency Coalition, told the committee that developers should disclose risks, independent auditors should verify those disclosures and the state should be notified quickly when something goes wrong. WITF reported that Wimmer characterized the bills as similar to laws enacted in California, New York and Illinois.

The committee discussion comes amid public concern about AI safety. WITF reported that a Quinnipiac University poll found 86% of Americans support requiring AI developers to meet independent safety standards, even if that slows development. Shusterman cited the same figure in her remarks.

The bills respond to a debate over how to address potential risks from advanced models, but they would not cover all AI systems. The coalition’s analysis says the proposals target large frontier developers, and Guzman acknowledged in his testimony that smaller models present issues beyond the bills’ reach. The supplied accounts do not report a committee vote or a timetable for further action.