A White House teleprompter operator used his position to win around $100,000 by placing bets on the prediction market Kalshi about what President Trump would say in his speeches, the company said on Thursday.

It was the latest accusation that someone had used inside information to make a profit on a prediction market like Kalshi and Polymarket, which have grown rapidly and transformed into cultural phenomena.

In this case, Kalshi said Gabriel Perez, a technical assistant to Mr. Trump, had placed wagers on common words that would appear in the president’s speeches, such as country names and economic terms. In March, when Kalshi’s surveillance systems flagged some of Mr. Perez’s trades, the company froze the funds in his account and referred the case to the Commodity Futures Trading Commission, the federal agency that regulates prediction markets.

Mr. Perez is in discussions with federal regulators to settle allegations that he used his inside knowledge of the president’s speeches to place the bets and has been cooperative, a person with knowledge of the talks said.

“We have charged this individual and have been assisting regulators on this matter and provided evidence we collected, as we do in any referral,” Robert DeNault, Kalshi’s head of enforcement, said in an emailed statement, referring to the company’s rules against insider trading.

A spokeswoman for the Commodity Futures Trading Commission said the agency could not “confirm or deny an investigation.” Mr. Perez could not immediately be reached for comment. ABC News earlier reported the incident.

Prediction markets allow people to wager on everything from state elections to World Cup matches to who will win on “Love Island.” The companies have attracted billions of dollars in trades, but their popularity has raised concerns about whether they violate state gambling laws and are being manipulated by those with insider information.

In April, three political candidates were found trying to bet on their own races. In June, federal authorities said they were investigating whether former Representative George Santos had made a bet about attending Mr. Trump’s State of the Union address.

In March, the White House warned staff against using insider information to place bets on prediction markets, just weeks before federal prosecutors charged an Army Special Forces soldier with using classified information to bet on the capture of Venezuela’s president, Nicolás Maduro, on Polymarket.

Mr. Trump is set to give a prime-time address on Thursday about election security and voting machines. On Kalshi, users are betting he will say “Save America Act,” “fraud,” and “Iran” more than three times, among other phrases and topics.

In a press briefing on Thursday, Karoline Leavitt, the White House press secretary, said that she had talked to Mr. Trump about reports that his teleprompter operator had been making money off his speeches on prediction markets and that he had called it a “disgrace.”

“This individual will no longer be here,” she said. “That was a decision by the president.”

Ms. Leavitt was also asked if other members of the White House staff were suspected of using inside information to make money on prediction markets. “Not to my knowledge,” she said.

She added, “There are very strict ethical requirements by all staffers and officials here at the White House” against such conduct.

Shawn McCreesh contributed reporting.

Kalshi, the start-up that lets users wager on almost anything, has struck a partnership with the artificial intelligence company OpenAI to show World Cup prediction market data in ChatGPT search results.

The tie-up, which has not been promoted publicly, uses Kalshi data to display the odds of victory in various matches in user queries in ChatGPT, search results show.

A ChatGPT search for “France and Spain,” for example, displays a graphic showing that the prediction market has assigned the French a 60 percent chance of winning on Tuesday, attributed to Kalshi data. A search about the countries playing in the other semifinal cites Kalshi data suggesting that England has a 55 percent chance of beating Argentina.

The deal with a prediction market is a first for OpenAI. For Kalshi, it is part of an ongoing campaign to integrate its wagering data far and wide to increase awareness and — ultimately — to bring more users to its site. Last year, Kalshi cemented a partnership to integrate its data across CNN and CNBC, including a real-time ticker powered by Kalshi data to run during news segments. In January, Polymarket, Kalshi’s largest rival, struck a similar deal to integrate its data across Dow Jones products, including The Wall Street Journal.

Regulators are facing increasing pressure to crack down on abuses in fast-growing and highly lucrative prediction markets. Correct bets on the timing of the U.S. military’s actions in Iran and Venezuela and company-related news have created growing concerns about leaks of sensitive information and potential insider trading. Earlier this year, a top federal financial agency opened an extensive investigation into Polymarket, raising inquiries about whether the site is was operating within the law.

OpenAI and Kalshi declined to comment.

In a recent update to its corporate help web page, OpenAI said that users “cannot place bets through ChatGPT,” and that the partnership appears to be limited to using Kalshi data only “for queries related to the 2026 World Cup.” The company also states that the data is “for informational purposes only.”

As prediction markets grow in popularity, more Silicon Valley companies are strategizing about how to get in on the action. Last year, Google struck deals with Polymarket and Kalshi to put prediction market results into Google Search results and on its Google Finance page.

And Mark Zuckerberg, chief executive of Meta, has directed lieutenants to build a stand-alone prediction markets app that would be directly integrated into Facebook and Messenger.

(The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. The two companies have denied those claims.)