NJ lawmakers weigh AI regulations amid industry and safety expert clash • New Jersey Monitor
New Jersey has discovered a bold new regulatory technology called 'asking firms to describe their own safety procedures', which the bill's drafters would like you to know is not the same as regulating anything. The package under consideration requires large AI developers to disclose how they handle frontier models, to the state attorney general, on paper. It is the legislative equivalent of asking the fox to submit a memo about the henhouse, and then filing it in a locked drawer marked 'audited'. The drawer, at least, is new.
The genuinely delicious part is the audit clause: the attorney general may inspect large AI firms directly or through private contractors, by which we mean the state will hire consultants to check whether Meta is behaving. Gentlemen, I give you the revolving door, now with a lanyard. You will not be shocked to learn who has thoughts on this. Meta public policy director Matt Lenz told lawmakers that large model creators — Meta and Anthropic both, thank you — should be shielded from liability for harms spurred by users. The model that might tell a teenager how to make something unpleasant is, in this telling, a public utility with an opinion column; the user, a rogue — shield the firm, blame the punter, repeat.
One notes, wearily, that New Jersey is weighing this while the actual frontier labs operate elsewhere, at speeds measured in weeks, with model cards that read like estate agent brochures. A disclosure regime keyed to what a company chooses to write down is not a leash; it's a stationery order. And the safety experts in the room, per the report, clashed with industry over all of it, which is the closest thing to accountability you get: a disagreement, minuted. Still, credit where due — somebody in the statehouse has at least noticed the machines exist, and Matt Lenz has turned up to make sure the noticing doesn't get too firm. The bill may pass; the liability shield, you'll find, will pass first.