By Benjamin Guggenheim

Not a subscriber? Sign up here to get this newsletter in your inbox.

• Exclusive: More congressional offices appear to be using AI to draft and edit remarks and floor speeches, according to a new comprehensive analysis by AI text-detection company Originality.ai. (Scroll for party breakdown and other details.)
• It’s unclear exactly how AI text-detection tools work, according to Hany Farid, a computer science professor at Dartmouth. Using AI for drafting and editing speeches and legislation holds both promise and peril.
• Lawmakers have introduced new proposals around AI governance and to liability around AI agents. The National Compute Grid launches to pool access to compute.

This is Benjamin Guggenheim. Welcome to WP Intelligence’s AI & Tech Brief, where we examine the transformative technology of artificial intelligence at the intersection of innovation, policy and power. Get in touch at Benjamin.Guggenheim@washpost.com or on Signal at ben_guggenheim.88.

The Lead Brief

Most organizations in the United States are experimenting with using AI in their workflows — and Congress is no exception.

According to a new analysis by Originality.ai, a Canada-based AI-text-detection company, a comprehensive evaluation of remarks on the Senate and House floors found that the percentage of lawmaker speeches classified as “likely AI” has been steadily increasing.

Of the 21,388 floor turns analyzed in 2025 and year-to-date in 2026, Originality.ai’s AI-detection tool found that the rate of “likely AI” speeches rose from 8.4 percent in 2025 to 12.2 percent in year-to-date 2026 (through Aug. 17).

According to the analysis, which I obtained exclusively, House lawmakers appeared to be more avid users of AI than members of the Senate. And Republican lawmakers in both chambers appeared to be heavier users than Democrats.

“They don’t surprise me at all,” Marci Harris, the co-founder and executive director of POPVOX Foundation, a nonprofit that helps governments incorporate emerging technology, said of the findings.

“That’s one of the first jobs that an intern gets trusted to work on. ‘Hi, you’ve been here for a week. Look at the way we do this in this office, and take the local newspaper and find interesting things that happen in the district and propose a one-minute speech,’” said Harris.

The analysis builds off an earlier report by the company that found congressional speeches classified as likely AI-generated have steadily increased in recent years. In 2020, before OpenAI released the first commercially available chatbot models, around 2 percent of floor speeches were classified as likely AI.

Zooming out: It’s worth noting that these AI-detection tools themselves have been the subject of some controversy.

Professors have been unleashing them on students to try to catch when their pupils use AI tools to write essays — though that has itself received pushback, with undergraduate and graduate students under review for academic dishonesty claiming that the AI-detection tools get it wrong.

“The problem with these techniques is you can’t really explain why. And the reason is they’re purely statistical. They’re like, ‘We’ve learned these patterns from AI. We’ve learned these patterns from humans. We don’t know what the patterns are,’” said Hany Farid, a computer science professor at Dartmouth who specializes in analyzing deepfakes.

According to my own review cross-referencing a randomly selected sample of floor turns analyzed by Originality.ai with Pangram, another AI-text-detection tool, the two tools classified most floor turns in similar ways and agreed on certain text being mostly AI-generated or not. But, in three cases, Pangram and Originality.ai had opposite conclusions of whether the given remark was human-generated or AI-generated.

Farid said that so much of writing these days can be a hybrid of AI-generated and also editing text and human text — and that people need to be careful of using AI-detection tools to accuse others of wrongdoing.

“It makes me uncomfortable as somebody who spent the last 30 years of [my life] thinking about forensic analyses,” said Farid.

Jon Gillham, the founder of Originality.ai, said in written commentary that he agrees with “much of what Farid shares,” and that one limitation with the tool is explaining how the detector produced its results.

“However, the trend of increasing AI use within politicians’ speeches in general, and likely heavy reliance among a few, is clear,” Gillham said.

U.S. Senate Majority Leader John Thune (R-SD) is trailed by a reporter as he walks past the Ohio Clock. (REUTERS/Jonathan Ernst/File Photo)

Legislative AI

Indeed, the report’s findings are broadly telling as to the increased use of AI tools by congressional offices — several of which acknowledged, in response to a request for comment about some of Originality.ai’s findings, that they do indeed employ AI tools to help draft and edit floor remarks. (You can look at a ranking of lawmakers by appearances of “likely AI” speeches in the linked report.)

“Our office, like most offices, utilizes House-authorized AI services such as [Microsoft] Copilot, which is now built into all our House systems, to optimize smaller daily tasks,” said Anthony Fakhoury, the communications director for Rep. Mariannette Miller-Meeks (R-Iowa). ”ALL written pieces of communications in our office go through several sets of eyes until it’s actually used.”

Scott Waters, communications director for Rep. Mike Lawler (R-New York), said that Lawler’s staff “responsibly uses AI in some cases to aid in drafting remarks.” The AI tools make the office more efficient,“ which is something all offices should aspire to,” Waters said.

However, some lawmakers have a different view on the use of AI, with Rep. Thomas Massie (R-Kentucky) taking to the House floor in September to criticize his colleagues for AI-drafted speeches.

“I think it was against the rules at one time to read your speeches in this body. Why is that? Because it was considered insincere,” said Massie. “This is not a robot speech, folks. I am not a robot congressman — never have been.”

Here are a few key takeaways from the Originality.ai report:

• The House saw significantly more “likely AI” speeches than the Senate in both years (11.4 percent in the House compared to 3.1 percent in the Senate in 2025).
• Republicans had higher rates of “likely AI” speeches in both chambers compared with Democrats. (In year-to-date 2026, the rate for House Democrats was 10.8 percent, while the rate for House Republicans was 19.3 percent.)
• There were some heavy users, with the tool finding that some lawmakers gave remarks that were “likely AI” for up to 90 percent of their speeches.
• Originality.ai classified a significant number of speeches (646) in 2025 and year-to-date 2026 as “likely AI,” with a confidence score of 100 percent.

“I think it’s reasonable for voters to want to understand how lawmakers and their offices are using [and] disclosing AI,” said Gillham. “Our own internal testing and independent studies show the tool is highly accurate but not perfect, and ideally suited for studies looking across a large dataset.”

Looking at a big chunk of the congressional remarks, many of them follow similar rubrics that highlight the work of constituents or champion projects built with federally appropriated funds in lawmakers’ respective districts.

So, it’s not necessarily surprising that congressional staffers would use AI, though Harris noted reports that staffers have used AI to draft legislation with errors — which can knot up the legislative process and, in the worst-case scenario, result in technical problems in the statute.

“The world is changing to keep pace with the executive branch, that is definitely using AI, to keep pace with their constituents that are sending them in massively longer letters, and lobbyists that are overloading them and increasing complexity of issues,” said Harris.

“So we are big fans of the use of AI on Capitol Hill, but definitely not unaware of the ways in which it needs to be handled with care and responsibly.”

The Legislative Chessboard

Lawmakers released a slew of new legislative proposals Wednesday on AI policy — as a coalition of cloud providers, universities, chipmakers and investors launched a new National Compute Grid to pool access to compute.

As I previewed in an interview with Anjney Midha, a leader of the effort, in an interview in August, the goal of the project is to create an accessible market for compute to small AI developers, government agencies and academic researchers alike.

The grid aims to leverage the massive amount of chip processing capacity that currently sits idle — permitting researchers to easily access compute at market prices reflecting live supply and demand through a scheduling system.

As part of the launch, National Compute, the new company operating the National Compute Grid, is donating $100 million in computing credits to the Department of Energy’s Genesis Mission to advance scientific innovation. The grid has already logged $5.1 billion in reservations, according to information received from the company.

As of Wednesday, users with a .gov and .mil email address will be able to join the grid through a technical preview phase.

New AI proposals: Maria Cantwell (D-Washington), the top Democrat on the Senate Commerce Committee, released a framework Wednesday outlining her principles for an AI framework. Cantwell says that the National Institute of Standards and Technology should develop clear standards for AI systems on catastrophic risk.

Cantwell furthermore says the government and independent experts should engage in constant oversight and stress-testing the models.

It’s a notable step from the Senate Democrat, who has privately objected to a different approach to AI regulation under development by Sen. Amy Klobuchar (D-Minnesota) and Senate Majority Leader John Thune (R-South Dakota).

“Benchmarks and the standards are preconditions for credible audits, and they do not yet exist,” Paul Lekas, executive vice president of global public policy and government affairs at the Software and Information Industry Association, said in a statement.

“Sen. Cantwell’s framework reflects core concepts that SIIA has long supported around risk management, governance, disclosure, incident reporting, resilience and more.”

Among other ideas to address “social harm” in children’s safety and the workforce, Cantwell says that the U.S. and China should establish a hotline to share information on major AI incidents.

Rep. Lori Trahan (D-Massachusetts) also unveiled new legislation Wednesday clarifying when courts can hold AI developers liable for actions by rogue agents. The bill establishes a cause of action in federal court. (As I explained in a recent edition, holding developers liable for harms by agents is tricky under current law.)

“When someone breaks the law and hurts you, you can take them to court. That shouldn’t change just because the wrongdoer is an AI agent,” said Trahan.

This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently from The Washington Post newsroom. Learn more about WP Intelligence.