Senate Drops Proposed State AI Moratorium from Budget Bill After Bipartisan Pushback and Procedural Objections

The unified budget bill being considered by Congress initially contained a provision that at first attracted little attention: a complete ban on states passing laws to regulate AI for ten years.  The proposed ban passed the House.  When the measure reached the Senate, it met with substantial pushback from both sides of the aisle.  Ultimately, the proposal was dropped from the Senate version of the bill, by an overnight vote on an amendment on July 1st.  

Opponents of the provision said that the bill would both hamstring states and leave Americans vulnerable to exploitation by AI; they also pointed out that Congress and the federal government are notoriously slow to catch up to technology – as demonstrated over and over regarding data privacy in the last decade.  Supporters of the ban said that allowing a patchwork of state regulations of AI (as has sprung up with privacy) would stifle innovation in the AI industry and slow the pace of its growth.  

It is not clear whether anyone will try to reinstate any AI limitations on the states, a move that was heavily favored by the tech industry.  Given that Congress is trying to pass the budget bill by July 4th, the prospects seem dim for the AI ban to be added back to the budget measure.  Time will tell whether Congress raises it in any future legislative efforts.  

WHY IT MATTERS

The proposal itself was noteworthy for at least two reasons: federalism/states' rights (it is rare for Congress to prohibit the states from passing laws generally, and especially on topics that Congress itself has not acted on yet); and the evolution of AI.   

With respect to federalism: it is very unusual for the federal government to outright prohibit the states from passing laws on something.  Although federal laws often state that they take primacy over state law on similar subjects, in practice what that often means is that federal law provides a baseline level of coverage that the states cannot alter.  States are often free to provide more protection than federal law provides, however.  It is even rarer for Congress to attempt to tie the states' hands on a subject as to which Congress itself has not even acted.  For the feds to say “you can't pass rules, even if we haven't tried” is not the usual model for relations between the federal and state governments.  Had this provision remained in the bill, we likely would have seen quite a lot of litigation challenging it on the grounds of states' rights.

The second area, AI safety, is a greater unknown. AI is still in its early stages.  Every day brings some weird headline about AI gone rogue.  The states and countries that so far have passed AI laws have tried to impose some guardrails that will at least protect humans from the most risky uses of AI.  If states were not free to do that – and if Congress remained true to its inability to regulate complex technology matters – it is not clear that industry would self-regulate against harmful practices.  Although the EU has passed a sweeping framework for regulation, the timing of its implementation is now in doubt; and it also does not apply to use of AI in the US.  Refusing to allow the states to act could have created a “wild west” atmosphere regarding AI.  

That is not to say that having multiple state laws, and no federal law, is a great model. It's not!  This is how the US operates with data privacy, and it makes compliance extremely difficult, especially for small companies.  But if the feds are unable to act, to strike a balance between consumer protection and AI innovation, it would seem to be an overreach for them to also require that the states sit on the sidelines.