Pa lawmakers push for AI safety rules | WITF

Pennsylvania has joined the grand American tradition of regulating the future one committee meeting at a time. On Monday the Pa. House Communication and Technology committee met, experts and lawmakers called for urgent action, and on the table sat HB2800 from Rep. Melissa Shusterman (D-Chester) and HB2829 from Rep. Manuel Guzman (D-Berks). Urgent, in this context, means a meeting has been held about bills that have been introduced. The frontier models in question are, at time of writing, being trained in other jurisdictions entirely, in buildings with better air conditioning and considerably better lawyers.

HB2800 would "establish measures to mitigate catastrophic risks from frontier AI models." Mitigate — not prevent, not license, not audit, not pause. It's the verb you borrow from insurers and weather forecasters, the linguistic equivalent of sandbags stacked at the door of a house already underwater. HB2829 would "set expectations for safety practices and incident reporting," which is the regulatory strategy of a parent who has worked out that expectations are far cheaper than consequences and never require a lawyer. Both are, in the strict sense, asking nicely. We watched the same earnestness in New Jersey a week ago; the models did not notice that either.

The article notes the federal government "has yet to provide a framework," the kindest available phrasing for a Congress that treats AI policy as a hot potato with a lobbying budget attached. So the states fill the vacuum: fifty definitions of "frontier," fifty thresholds, fifty incident forms, and a company operating everywhere collects fifty slightly different letters to file and ignore. The catastrophic scenarios the bill titles reach for are not state-shaped at all; they are compute-shaped, and compute has never respected a county line, let alone the Chester one. Still, the meeting happened, urgent action was requested, and somebody has to write the minutes.