Congress Faces an AI Regulation Gap of Its Own
As Washington debates guardrails for frontier models, many lawmakers acknowledge limited hands-on AI experience—raising questions about policy capacity.
Key facts
- More than two dozen lawmakers and governors told Axios they do not use AI or use it rarely.[1]
- Legislative staff AI awareness and office policies have increased in 2026, according to the NCSL survey.[2]
- Congress faces disagreement over federal AI safeguards, timing and the role of government oversight.[3]
Washington’s AI debate is running into an awkward credibility problem: many of the lawmakers under pressure to regulate frontier systems say they rarely use artificial intelligence themselves.
An Axios report published September 16 found that more than two dozen members of Congress and governors either do not use AI or use it only occasionally. Some lawmakers defended that distance, arguing that legislators do not need to use a technology personally to write rules for it. But the admission arrives as Congress considers questions involving model evaluations, catastrophic-risk thresholds, independent audits, national security and liability.
What changed
The immediate development is not a new federal law, but a sharper public test of whether Congress understands the systems it may soon regulate. The disclosure follows several days of heightened political attention after AI executives and researchers issued unusually direct warnings about the pace and risks of development.
The political response remains fragmented. Some senators are pursuing bipartisan legislation that would require stronger safeguards and outside assessments. Other leaders, including House Speaker Mike Johnson, have signaled that Congress should not rush to lead on AI safety. President Donald Trump has rejected calls for a broad slowdown and characterized some warnings as exaggerated.
That combination—rising urgency, limited technical familiarity and no clear legislative timetable—creates a policy gap that the industry may try to fill through voluntary standards or self-regulation.
Why it matters
AI policy is unusually difficult because the technology is both highly technical and rapidly changing. Lawmakers can rely on expert agencies, hearings and outside advisers, but those mechanisms work only if policymakers can distinguish meaningful evidence from vendor messaging. Limited direct experience does not automatically disqualify a legislator; it does increase the importance of independent testing, transparent documentation and technical staff who are not dependent on the companies they oversee.
There is also a democratic risk. If elected officials cannot confidently evaluate basic claims about model capabilities, the practical rulebook may be written by laboratories, consultants and lobbyists with commercial interests. That could produce weak safeguards—or rules so broad and poorly targeted that they impede useful applications without addressing the most serious hazards.
A 2026 survey from the National Conference of State Legislatures offers a more nuanced picture. Nearly 70% of surveyed legislative staff said they were aware of AI use in other offices, and roughly half reported that their own office had an AI policy. The staff-level adoption suggests government is not uniformly disconnected from the technology, even if many elected officials are.
What remains uncertain
The central question is whether Congress can convert concern into durable oversight before the next election cycle and the next generation of models. Bipartisan interest is growing, but agreement on enforcement, federal authority and the definition of a dangerous system remains elusive.
For now, the experience gap is less a story about lawmakers being incapable of regulating AI than a warning about institutional readiness. The first serious test may not be whether Congress can pass a sweeping AI law, but whether it can build the technical capacity to know which rules deserve to survive.