AI Policy After the Midterms: What Can the Next Congress Actually Change?
A different majority can change the agenda. It cannot turn a proposal into law by itself.
By Theoplis Stewart II
Reporting snapshot: November 10, 2026
Carol M. Highsmith / Library of Congress, Prints and Photographs Division · No known restrictions on publication
The next argument over artificial intelligence will take place in a different Congress. The rules governing AI have not changed simply because the election did.
As of November 10, The Associated Press projects Democratic majorities in both the House and Senate, according to CQ Roll Call’s election tracker. Those are news-agency projections, not a declaration that every race has been certified. The incoming congressional terms begin January 3, 2027. The current Congress remains in office during the intervening period. [1] [2] [16]
That distinction matters to anyone trying to understand what comes next for AI. A new majority can change which proposals receive attention and which companies or agencies face questioning. Passing legislation is a separate process. So is implementing it.
For journalists, creators and businesses using AI, the useful question is not whether a “blue wave” guarantees more regulation. It is what specific measures would do, how far they have advanced, and who would actually be covered.
First, what is actually up for a vote?
The House’s published list of legislation for the week of November 9, updated November 10, does not include the principal AI proposals examined below. It includes measures concerning nuclear licensing, small businesses, veterans and retirement plans, among other subjects. That is a dated schedule check, not a claim that Congress cannot add items or consider other legislation relevant to AI. [3]
An introduced bill is not a scheduled vote. A committee’s approval is not passage by the full chamber. Placement on the Senate calendar does not establish a date for consideration. And a discussion draft is an earlier step still. The distinction is essential when a policy announcement is described as a coming legal obligation. [11] [12] [13] [17]
The central finding: the election changes the opportunity to advance an AI agenda. It does not establish that the proposals in this article are about to pass.
Where the incoming agenda points
Reuters reported November 6 that Democrats were preparing measures addressing electricity costs associated with data centers, independent evaluations of advanced AI systems, and liability when AI products cause harm. Those are reported priorities, not an enacted package or a confirmed voting timetable. [4]
There is a practical mechanism behind an agenda shift. House majority leadership has considerable influence over floor consideration, while committees develop legislation and oversee implementation. A different majority can redirect that work without every member agreeing on a single AI philosophy. [18]
But the policy dispute is not simply “Democrats favor safeguards; Republicans oppose them.” The March 2026 White House legislative framework also calls for protecting children, addressing AI-enabled scams, respecting creators’ rights and preventing ordinary electricity customers from bearing data-center costs. It favors a national approach intended to avoid conflicting state requirements. These are the Trump administration’s recommendations to Congress, not legislation enacted by the announcement. [5]
There is overlap in stated objectives. The substantive questions concern the mechanism: mandatory requirements or voluntary commitments; federal uniformity or preserved state remedies; public enforcement or private lawsuits; and which companies must meet which standards. Agreement that a problem exists does not establish agreement on a bill.
“Frontier supercomputer” — Oak Ridge National Laboratory · CC BY 2.0 · Resized and JPEG-recompressed.
Five proposals, five status checks
The following is a selected federal-policy watchlist, not an inventory of all AI legislation. Dates identify the latest actions located in the records reviewed for this article.
FRONTIER Act, H.R. 9925
What the proposal addresses: Oversight of advanced AI development, with obligations differentiated by developer scale.
Verified stage: Introduced July 23, 2026; referred to House Energy and Commerce and Science, Space, and Technology. No passage established. [6] [7]
Algorithmic Accountability Act of 2025, H.R. 5511
What the proposal addresses: Impact assessments for covered entities using automated systems in critical decisions.
Verified stage: Introduced September 19, 2025; referred to House Energy and Commerce. [9]
Artificial Intelligence Civil Rights Act of 2025, H.R. 6356
What the proposal addresses: Discrimination safeguards and evaluations for covered consequential algorithmic decisions.
Verified stage: Introduced December 2, 2025; referred to House Energy and Commerce and Oversight and Government Reform. [10]
NO FAKES Act of 2026, S. 4591
What the proposal addresses: Federal protection concerning unauthorized digital replicas of a person’s voice or visual likeness.
Verified stage: Reported with an amendment June 24, 2026; placed on the Senate calendar. Not an enacted law in the reviewed record. [11] [12]
CLAIM Act discussion draft
What the proposal addresses: Developer liability for harmful AI-agent conduct under the proposed framework.
Verified stage: Released by Rep. Lori Trahan on October 7, 2026 for feedback. No introduced bill number or floor-vote date was established in this review. [13]
Frontier-model oversight is not a rule for every AI user
Rep. Jay Obernolte, a California Republican, introduced the FRONTIER Act with Democratic Rep. Lori Trahan of Massachusetts and other bipartisan supporters. Its authors describe a national framework combining advanced-model oversight with continued development. [8]
The introduced text differentiates obligations for large and very large frontier developers. Its provisions include transparency, risk-management frameworks, independent verification and incident reporting. That is materially different from requiring every photographer, newsletter publisher or small studio using a chatbot to obtain a federal AI license. [7]
The important questions are which developer meets the definitions, what evidence an evaluation must produce, and how an incident triggers a duty. A future version could change those details; the introduced text is a proposal, not an instruction already binding every business.
Decision-making tools raise different questions
H.R. 5511 and H.R. 6356 focus on covered automated or algorithmic systems and consequential decisions, rather than treating all AI-assisted activity as identical. Their assessment and civil-rights provisions should be read alongside their definitions and coverage limits. [9] [10]
For a media business, drafting a caption and using an algorithm to influence a hiring decision are therefore different situations to investigate. That comparison is an editorial illustration, not a legal finding that either bill applies to a particular company. Coverage would depend on the enacted language, if any, and the actual use.
A digital replica is not the same issue as model training
The NO FAKES proposal concerns rights in an individual’s voice and visual likeness. The reported Senate version is a concrete proposal about digital replicas, not a general ruling that all copyrighted training material must be licensed. Its calendar position demonstrates legislative movement without establishing final passage. [11] [12]
That difference matters to a performer whose voice is copied, a photographer licensing an image, and a reporter documenting an impersonation. Their interests may intersect, but they are not interchangeable legal questions. Consent, permitted uses, exceptions and remedies must be examined in the operative text rather than inferred from the bill’s name.
Agent liability remains a proposal
Trahan’s October discussion draft would make developers liable under specified circumstances when an agent injures a third party through conduct that would be unlawful if performed by a person. Her announcement describes exceptions involving intentional or negligent conduct by a user or a company that modified the system. It also describes preservation of state law. [13]
That is not evidence that Congress has already assigned liability for every agent failure. Nor does it establish that a publisher or business owner could delegate responsibility without consequence. Those conclusions would require reading the eventual law, its exceptions and the facts of a particular case.
What remains in place
The incoming Congress does not replace the president. Donald Trump’s administration remains a separate participant in the policy process. Its July 2025 AI Action Plan emphasizes innovation, infrastructure and international security; its March 2026 legislative framework sets out proposals for Congress. An election result does not, by itself, repeal those executive policies or enact their recommended legislation. [5] [20]
There is also existing federal law, not merely competing proposals. The TAKE IT DOWN Act became Public Law 119-12 on May 19, 2025. Sponsored by Sen. Ted Cruz, it passed the Senate by unanimous consent and the House 409–2. It addresses nonconsensual intimate imagery, including qualifying computer-generated depictions. It is not a general labeling law for every AI-generated photograph or news article. [14]
That example is important for assessing the political direction: a targeted protection obtained broad agreement before these midterms. It does not guarantee comparable agreement on model audits, private liability or the relationship between federal and state law.
The limits of a House majority
A House majority can advance a bill through its chamber. Enactment ordinarily requires both chambers to agree on identical legislation and present it to the president. A presidential veto can be overridden by two-thirds of the members present and voting in each chamber, provided a quorum is present. That threshold is different from an ordinary majority. [16] [17]
The Senate presents another distinction. Final passage usually requires a simple majority, but ending debate on ordinary legislation through cloture generally requires three-fifths of senators duly chosen and sworn, normally 60. Not every measure faces that route, and “every bill needs 60 votes” would be inaccurate. [15]
There is a further deadline. Bills that do not become law by the end of the 119th Congress do not automatically carry over as pending bills in the 120th. They must be introduced again. Language, sponsors and bill numbers can change. A 2026 proposal is useful evidence of an agenda, but not a substitute for the text that lawmakers may consider in 2027. [19]
These constraints do not make an election irrelevant. They explain where its effects can first appear: committee priorities, hearings, requests for records, negotiations and legislative text. None should be confused with an obligation that has already taken effect. [18]
What to watch instead of a partisan label
For a journalist or creator, three checks make this debate more usable.
Read the coverage definition. A frontier-model developer, a distributor of digital replicas and an employer using a decision tool occupy different positions. A headline saying “Congress regulates AI” does not establish which one faces a duty. [7] [9] [10] [11]
Separate a proposal’s goal from its result. Protecting ratepayers is a stated objective in both the reported Democratic agenda and the White House framework. Whether any eventual measure changes electricity bills requires examining its design and implementation. It does not follow that an AI subscription will become cheaper. [4] [5]
Follow the version and the vote. An amended committee text can differ from the introduced bill. A vote to advance a measure is not always a vote to pass it. The reliable trail is the official record: text, action, chamber decision, presidential action and effective date. [17]
The question after the midterms is not which party can be assigned a permanent label of “pro-AI” or “anti-AI.” It is which coalition can turn a defined proposal into an enforceable rule, and what that rule would actually require.
The election changes who can set the agenda. The legislative record shows whether the rules change.
Sources and reporting note
This is document-based policy reporting, prepared with AI assistance and checked against official legislative records, published administration proposals and attributed news reporting. No original interviews were conducted. The election baseline uses AP projections as reported by CQ Roll Call, rather than asserting completed certification. Legislative status and schedules are a November 10, 2026 snapshot; they can change. The watchlist is not a declaration that votes are imminent and is not a comprehensive state-law or compliance guide. Company-specific legal decisions require advice from a qualified attorney.
[4] Reuters, Democrats planning flood of AI bills when they take control of Congress, November 6, 2026
[18] Congressional Research Service, Introduction to the Legislative Process; public HTML reproduction
Comments