White House urges Congress to take a light touch on AI regulations in
new legislative blueprint
[March 21, 2026]
By SEUNG MIN KIM and MATT O'BRIEN
WASHINGTON (AP) — The White House said Friday that Congress should
“preempt state AI laws” that it views as too burdensome, laying out a
broad framework for how it wants Congress to address concerns about
artificial intelligence without curbing growth or innovation in the
sector.
The legislative blueprint outlines a half-dozen guiding principles for
lawmakers, focusing on protecting children, preventing electricity costs
from surging, respecting intellectual property rights, preventing
censorship and educating Americans on using the technology.
House Republican leaders swiftly endorsed the framework and said they're
ready to work “across the aisle” to pass legislation, but doing so would
be a heavy lift, requiring agreement with Democrats in the Senate as
public divisions over AI run deep.
The announcement comes as state governments have forged ahead on their
own regulations for AI while civil liberties and consumer rights groups
lobby for more regulations on the powerful technology. The industry and
the White House have pushed back, arguing that a patchwork of rules
would hurt growth. Trump signed an executive order in December to block
states from crafting their own regulations.
“This was in response to a growing patchwork of 50 different state
regulatory regimes that threaten to stifle innovation and jeopardize
America’s lead in the AI race,” said White House AI czar David Sacks in
a social media post Friday.
Sacks said the next step is to work with Congress to turn the
administration's principles into federal legislation.
AI legislation could need bipartisan support to pass
While passing sweeping AI legislation will be difficult, especially in a
midterm election year, the framework appeared designed to appeal to some
AI-wary Republicans and Democrats with a focus on widespread and
bipartisan concerns, such as the harms that AI chatbot companionship can
pose to children and the electricity costs of AI infrastructure.
“It covers basically all the key sticking points I think that might stop
an AI bill from moving through Congress,” said Neil Chilson, a
Republican former chief technologist for the Federal Trade Commission
who now leads AI policy at the Abundance Institute. “It reads to me as
an attempt to build a larger tent, even if it doesn’t give everybody
everything that they want.”
But it has already been panned by some Democrats, including U.S. Rep.
Josh Gottheimer of New Jersey, who said in a statement it “fails to
address key issues, including strong accountability for AI companies,
under the guise of protecting children, communities, and creators.
Americans need protection — but this means nothing if we allow the AI
industry to be the Wild West.”
Whether AI legislation can pass both chambers of Congress could also
rely heavily on the support of Republicans like U.S. Sen. Marsha
Blackburn of Tennessee, who has introduced her own AI bill, and last
year was instrumental in thwarting Trump's earlier attempt to deter
state governments from regulating AI. Blackburn on Friday called Trump's
framework a roadmap and welcomed the administration to the “important
discussion" of getting a bill passed.
States already regulating AI don't want to be preempted
Several states — including California, Colorado, Texas and Utah — have
already passed laws that set some rules for AI across the private
sector.
With bipartisan support in the Texas legislature, a new AI law that took
effect this year in the Republican-led state requires government
agencies and health care providers to disclose when they are using AI to
interact with consumers or answer questions. The law also prohibits the
development of AI that encourages a person to commit suicide, harm
themselves, harm another person or engage in criminal activity.
A federal law following Trump's framework “could knock out parts of
Texas’s AI code while leaving some parts standing,” said Saurabh
Vishnubhakat, a professor at Yeshiva University’s Cardozo School of Law.
“The fact that it’s a Republican governor I don’t think is going to save
Texas’ law from preemption.”
Also vulnerable is Colorado’s law, which is aimed at preventing AI from
discriminating against people when making consequential decisions about
things like hiring and medical care. It was passed in 2024 but isn't set
to take effect until later this year. Lawmakers hope to rework the
regulations before then.
[to top of second column]
|
President Donald Trump speaks at a dinner with Japan's Prime
Minister Sanae Takaichi in the State Dining Room of the White House,
Thursday, March 19, 2026, in Washington. (AP Photo/Julia Demaree
Nikhinson)
Colorado State Rep. Jennifer Bacon, a Democrat, said voters don’t
want to stifle innovation or fall behind China “but our constituents
are interested in not becoming China.”
California's Democratic Gov. Gavin Newsom has vetoed some AI bills
while signing into law others. His office criticized Trump's
framework Friday.
“Yet again, Donald Trump is trying to gut laws in California that
keep our residents safe and protect consumers — a core state
responsibility,” Newsom’s spokesperson Marissa Saldivar said in a
statement.
The Trump administration says it doesn't think Congress should
preempt all state regulatory powers over AI, including enforcement
of general laws against AI developers, “to protect children, prevent
fraud, and protect consumers.” It also says Congress shouldn't
interfere with local authorities in deciding where to place data
centers and other AI infrastructure, or how states procure their own
AI tools for law enforcement or education.
However, it says states “should not be permitted to regulate AI
development,” shouldn't penalize AI developers for a third party's
unlawful conduct using their product, and “should not unduly burden
Americans’ use of AI for activity that would be lawful if performed
without AI.”
Trump's AI proposal appeals to concerns about data centers,
copyright
As backlash against data centers has increased along with rising
power prices, the White House had previously stepped up pressure on
AI companies and the power sector to do more to address the issue --
including having AI companies sign voluntary pledges earlier this
month to build their own power generation plants.
Some AI safety advocates are pushing for Blackburn and other
influential Republicans to insist on more protections against AI's
most catastrophic risks to national security or the economy, such as
out-of-control AI agents or the widespread replacement of human
workers.
“We have companies that explicitly are hoping to replace human
labor," said Brendan Steinhauser, a former Republican strategist who
now leads The Alliance for Secure AI and believes Trump's framework
doesn't do enough to address risks. "Tinkering at the edges with
upskilling and job training is just not going to make an impact on
that. I just don’t think we as a country are taking this seriously
enough.”
The framework aims for a more balanced approach to another
controversial topic: AI and copyright.
It recommends against wading into the legal fights between artists
and creators and the technology companies that have ingested huge
amounts of copyrighted works to build AI systems that can generate
new text, images and sound.
The Trump administration “believes that training of AI models on
copyrighted material does not violate copyright laws,” according to
the document, but acknowledges “arguments to the contrary exist and
therefore supports allowing the Courts to resolve this issue.”
That language was welcomed by trade group AI Progress — a coalition
that includes Amazon, Anthropic, Google, Meta, Microsoft, Midjourney
and OpenAI.
Tech companies have been fighting dozens of copyright infringement
lawsuits from writers and publishers, visual artists, music record
labels and others. Judges have largely sided with AI developers in
allowing for the “fair use” of copyrighted works to create something
new, but some have questioned how the materials were obtained. A
federal judge in September approved a $1.5 billion settlement
between Anthropic and authors who allege nearly half a million books
had been illegally pirated to train its chatbot.
—
O'Brien reported from Providence, Rhode Island. Contributing to this
report were AP writers Colleen Slevin in Denver, Trân Nguyễn in
Sacramento, California, and John Hanna in Topeka, Kansas.
All contents © copyright 2026 Associated Press. All rights reserved |
