Federal AI Preemption: What the Blackburn Draft Would and Wouldn't Do

This post is informational and reflects a legislative proposal as of March 2026. It analyzes a discussion draft, not enacted law, and the underlying negotiation is ongoing. For advice on your situation, consult a licensed attorney.

In March 2026, Senator Marsha Blackburn released a nearly 300-page discussion draft of a national AI policy framework, answering the White House’s call for federal preemption of state AI laws. The headline is preemption. The part worth reading is the fine print, because a preemption bill is defined by what it carves out.

For companies watching the state-by-state patchwork with hope that one federal standard will replace it, the draft is a caution: even if something like it passes, a meaningful amount of state exposure would survive. Three carve-outs stood out in the draft.

State attorneys general keep child-protection enforcement

The draft preserved state AG authority to enforce child-protection measures. For any company whose product reaches minors, that means the most active enforcers in the privacy space would not be displaced on the issue they care about most.

Generally applicable consumer-protection and product-liability claims survive

The draft’s structure preserved generally applicable state consumer-protection and product-liability claims. Preemption aimed at AI-specific statutes does not sweep away the ordinary state-law theories a plaintiff can already bring, which is where a great deal of real-world exposure lives.

The copyright provision cuts the other way

Separately, the draft stated flatly that training AI on copyrighted works is not fair use. That is not a preemption point; it is a substantive position that would raise, not lower, exposure for model developers and some downstream users.

The takeaway

The instinct to treat federal preemption as the end of state AI compliance is the wrong read of this draft. State enforcement survives it, particularly on children’s data and through generally applicable consumer-protection law. Planning your compliance program around a single federal standard that has not arrived, and that would leave these channels open even if it did, is a bet against the way the draft is actually written.

This remains a proposal in an active negotiation, so treat the specifics as a moving target. The durable point is the framing: read the carve-outs, not the headline. (For how this fits with AI governance and privacy compliance, see the firm’s AI Governance and Privacy & Data Security practices.)

If you are building AI compliance around what federal law might do, talk to MB Legal about the exposure that survives either way.

This post is informational and does not constitute legal advice. It describes a legislative discussion draft, not enacted law, and the proposal continues to change. For guidance on your situation, consult a licensed attorney.