The GAAIA Stakeholder Map: Who Wins, Who Loses, and What Each Outcome Requires of Compliance Teams

A 269-page bill arrived in Washington on June 4. It has two sponsors. It has no committee
referral yet. And it may be the most consequential federal AI document in years, or it may
never move.

Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released the discussion draft of the Great American Artificial Intelligence Act of
2026 on June 4. At its core, the bill proposes a single, specific trade: the federal
government takes control of how AI systems are *built*, and states keep control of how those
systems are *used*. That development-vs.-use line is precise by design. Understanding it is the
prerequisite for understanding what every stakeholder in this fight is actually fighting for.

—

The Preemption Mechanic

The proposed three-year moratorium applies specifically to state and local laws that regulate AI
development. Under the Broadband Breakfast analysis of the draft,
states would lose the ability to legislate on how AI systems are built. They would retain
authority over how those systems are deployed and used within their borders.

The practical meaning: a state law requiring frontier labs to conduct safety evaluations during
model training, that’s development regulation. Frozen for three years under this draft. A state
law prohibiting an employer from using an AI hiring tool that produces discriminatory outcomes –
that’s use regulation. Untouched.

The distinction isn’t arbitrary. It reflects a theory about where regulatory fragmentation
causes the most harm to AI development. Federal sponsors argue that fifty different state
frameworks for how models must be trained, tested, and documented creates compliance costs that
favor incumbents and block new entrants. The three-year window is framed as a breathing room
for federal standards to catch up with the technology.

The draft also requires frontier labs to submit their models for mandatory federal testing and
evaluation. That’s a harder ask than the voluntary 30-day pre-release window in the June 2
Executive Order. Voluntary and mandatory are not interchangeable. Compliance teams at frontier
labs who processed the EO as a one-page action item now have a second, more demanding framework
to assess, one that isn’t law yet but signals where congressional intent is pointing.

—

The Four Stakeholder Positions

Four distinct groups have incompatible interests in this bill’s outcome. Their positions aren’t
just rhetorical, each translates into specific compliance, political, or business consequences.

*Federal sponsors.* Obernolte and Trahan have built a bipartisan framing around a
competitiveness argument: without federal preemption of development regulation, American AI
development will be slowed by a patchwork of state requirements that European and Chinese
competitors don’t face. The mandatory frontier lab vetting provision gives the federal framework
teeth, it’s not just preemption of state rules, it’s replacement with a federal testing regime. Their coalition depends on keeping frontier labs neutral or supportive on the vetting provision
while building enough cross-aisle support to survive committee.

*Frontier labs.* The mandatory vetting requirement is the variable. Preemption of state
development laws is, from a pure compliance cost perspective, attractive to large AI developers
who currently face overlapping state requirements. But mandatory federal model submission is a
different calculation. It creates disclosure obligations, potential classification of training
data as subject to federal review, and a government touchpoint on every major model release. Whether the largest labs read this as acceptable regulatory overhead or as a threat to
proprietary development processes will determine their lobbying posture, and their posture will
determine whether the bill has industry cover in Congress.

*Consumer and civil society advocates.* Public Citizen has
argued directly that the bill strips states of authority to protect consumers from
algorithmic bias, fraud, deepfakes, and employment discrimination without providing a federal
guarantee of equivalent protections. This is advocacy framing, not a neutral legal reading –
but the structural concern is real. A three-year moratorium on development regulation, paired
with slow-moving federal rulemaking, creates a protection gap. States that have enacted
consumer-facing AI protections framed as development requirements would see those laws
suspended. If federal replacements don’t materialize within the moratorium window, consumers
lose ground. That’s the coalition Public Citizen is building around: not opposition to federal
AI regulation per se, but opposition to preemption without a guarantee.

*State enforcement regimes.* Three states have enacted AI laws that sit directly in the
moratorium’s path.

Illinois SB 315 requires annual algorithmic audits for employers using AI in hiring. Connecticut
PA 26-15 imposes requirements on AI companion app developers. Colorado SB 26-189 addresses
algorithmic discrimination. Each of these includes development-side requirements. Under the
draft’s moratorium, enforcement of those requirements, to the extent they touch development –
would be suspended for three years. State AGs who championed these laws don’t have obvious
incentives to support a bill that suspends their authority, regardless of what federal
replacements are promised.

The political geography here matters: Colorado, Connecticut, and Illinois are not states where
congressional Republicans hold significant leverage. Trahan’s involvement as a Democratic
co-sponsor is the attempt to manage that problem. Whether it’s sufficient is an open question.

—

The EO and the GAAIA: Complementary or in Tension?

The June 2 Executive Order and the June 4 discussion draft arrived two days apart and address
overlapping territory. They aren’t in direct conflict, but they aren’t fully complementary
either.

The EO establishes a voluntary 30-day pre-release window for frontier AI models and mandates
DHS Binding Operational Directives within 30 days for federal agencies. It’s an executive
action, it operates within existing agency authority and doesn’t require congressional approval. The GAAIA is legislation, it requires passage, signature, and would create new statutory
authority that the EO cannot.

Don’t expect the two to function as a unified policy framework. The EO moves fast and operates
on existing authority. The GAAIA, if it advances, would layer a mandatory statutory framework
on top of, and potentially in tension with, the voluntary EO mechanisms. A frontier lab that
participates in the EO’s voluntary 30-day review process may still be subject to the GAAIA’s
mandatory vetting requirements if the bill passes in its current form. Compliance teams
shouldn’t read the EO’s voluntariness as a signal about the GAAIA’s direction. They’re moving
on separate tracks.

—

The State Law Exposure Map

Three specific state laws define the immediate compliance exposure for companies asking whether
the GAAIA moratorium would relieve their current obligations.

The short answer: not yet, and possibly not at all.

Illinois SB 315 annual audit requirements, Connecticut PA 26-15 developer requirements, and
Colorado SB 26-189 algorithmic discrimination provisions are active law today. The GAAIA is a
discussion draft. The moratorium doesn’t exist until the bill is passed and signed. Companies
that pause compliance programs on the assumption that the moratorium will pass are making a
regulatory bet with real legal exposure on the downside.

The more useful framing: use the GAAIA’s development-vs.-use distinction as an analytical tool
for auditing your current state law compliance stack. For each state law you’re operating under,
ask whether the requirement in question is a development regulation or a use regulation. That
analysis is worth doing regardless of the GAAIA’s legislative fate, because it maps your
regulatory exposure across both the current state-law landscape and the federal framework that
may replace it.

—

What Compliance Teams Should Do Now

Four actions are warranted at the discussion draft stage.

First, track the bill’s status. Formal introduction and committee referral are the first real
signals of institutional support. A discussion draft that doesn’t get introduced within 60 days
is almost certainly stalled.

Second, audit your state law obligations against the development-vs.-use line. Identify which
of your current compliance requirements would be affected by a moratorium and which would
survive it. That’s a useful exercise for scenario planning regardless of the bill’s outcome.

Third, monitor frontier lab positioning on the mandatory vetting provision. If large AI
developers break against the bill over the disclosure requirements, the coalition fractures and
the bill’s chances narrow significantly.

Fourth, don’t treat the EO and the GAAIA as the same thing. Voluntary federal access under the
EO and mandatory federal vetting under the GAAIA carry different compliance obligations. The federal-vs.-state AI compliance question has been building across
multiple cycles. This bill is the most concrete legislative expression of it yet, which means
the analytical framework your team develops now will stay useful as this fight continues.

TJS synthesis: The GAAIA’s two-year preemption fight is still years from resolution, if it
resolves at all. But the bill has already done one thing: it’s forced a precise articulation of
the development-vs.-use line that will shape every subsequent federal and state AI legislative
debate, regardless of whether this specific draft ever passes. Compliance teams that internalize
that distinction now will have a durable analytical advantage as the regulatory landscape
continues to fragment. The mandatory frontier lab vetting provision is the provision most likely
to reshape the bill before it’s introduced, watch whether frontier labs lobby against it, and
watch whether Obernolte and Trahan are willing to negotiate it down. That negotiation will
reveal whose interests the bill is actually designed to serve.