Good morning! I hope this morning dispatch, focused on the latest updates about ongoing TPM storylines, has been useful. Starting next week, TPM members will be able to recieve it as an email. If you’re not a member, consider becoming one!

Propaganda Meets Reality

Ahead of President Trump’s big speech last month touting supposedly new information about election interference (there was little new info, and no new conclusions to draw from it), the Department of Homeland Security claimed to have uncovered “tens of thousands of non-citizens who are illegally on the states’ voter rolls.” That press release inspired credulous headlines in major news outlets, even as experts scrambled to explain that its conclusions appeared to be baseless, and likely founded on a database that would not allow the department to make such claims.

Now, despite DHS’ July assertion that it found “15,903 non-citizens registered to vote in Nevada,” the department has privately told state election administrators that it has identified on the voter rolls only a fraction of that number of “potential” noncitizens — 185, according to emails and recordings obtained by the New York Times.

Whats more, the administration has not given Nevada actionable information about who those 185 are — it hasn’t even provided names, per the Times. It also has apparently not alleged any of them voted. “In the virtual meeting, federal officials told Nevada that the nearly 16,000 they had originally identified represented ‘the ceiling’ of potential noncitizens,” the Times reports.

21-Year-Old Uses AI to ‘Test’ a ‘Theory.’ Chaos Ensues.

Following some impressive digital sleuthing, the Guardian identified and spoke with a 21-year-old who appears to be behind the website that put out fake polling data about California, Nevada and the gubernatorial primary in Wisconsin that saw David Crowley defeat Francesca Hong. The Hong polling, in particular — it showed her 20 points ahead, only for her to later lose her election — helped launch a thousand thinkpieces.

• Rahil Prakash, the guy the Guardian identified, told the newspaper the polling miss in the Michigan governor primary, when surveys showed Abdul El-Sayed with more of an advantage over Haley Stevens than he actually had on primary day, gave him the idea for the “experiment.” “This is what inspired me to test my theory,” said Prakash. “I was a bit shocked, perhaps at how fast this sort of information can spread.”
• Prakash said he used AI to create the fake polling results.
• Speculation has abounded that the whole operation was some kind of scheme to make money on betting sites such as Kalshi or Polymarket, even though “Median Strategies,” the outfit behind the “poll,” insisted the “project was not undertaken for financial gain or to influence the value of any financial or prediction-market position.” It has “all the earmarks” of a “pump-and-dump scheme,” American Prospect editor Dave Dayen wrote. “I think this falls into the category of hard to prove but very hard to think it’s anything else,” Corey Frayer, a former financial regulatory staffer in Congress now with the watchdog group Consumer Federation of America, told Dayen.
• Nonetheless, Prakash denied to the Guardian he placed bets on any prediction markets amid his scheme. “No, not at all,” he said.
• If this is the full story, it’s a wild cautionary tale about AI and our current media environment, where slop easily insinuates itself into our social media feeds, and even into coverage by major news outlets and statements by political campaigns. It also may not be the full story.

Tabs

• The Daily Beast has published what it says are two letters from Natalie Harp to Trump. They are bizarre!

“And please, when I fail, will you tell me? You have the absolute right to cuss me out, if need be, when I deserve it, because no one knows or cares about me more.”…“With all my heart, Natalie”

— Bill Grueskin (@bgrueskin.bsky.social) 2026-08-21T00:11:50.088Z

• Mike Rogers, the candidate running for Senate against Abdul El-Sayed in Michigan, backed a one-year moratorium on data centers, The Detroit News reports — a clever finger to the wind amid widespread distrust of AI and the facilities needed to support it.
• Relatedly, a Reuters analysis of money in this year’s election finds that “Crypto companies, artificial intelligence firms and online betting have emerged as the biggest industries shaping the 2026 midterm races, driving a record amount of business spending on congressional contests, according to campaign finance watchdogs and interviews with more than a dozen political ​strategists from both parties.”

Man of the Hour

Its Barron Trump, the college sophomore who is worth $150 million. Much of that fortune comes from his family’s various cryptocurrency schemes, Forbes suggests.

Are We at War?

Oman and the UAE are thrilled that parts of their countries are now “U.S. Territory.”

President Donald Trump lost the popular vote in the 2020 election by 4.4%. That hasn’t stopped him, though, from latching on to a highly unusual Census Bureau report published Tuesday afternoon that allegedly found less than two-hundredths of a percent of voters in that presidential election were noncitizens.

According to the report, 24,000 out of 128 million voters in the 2020 presidential election were ineligible “non-citizen” voters. Unnamed authors of the seven-page report, partially titled “A Beginning Analysis,” said the bureau still has to analyze 32 million voter records. According to Pew, some 158.4 million ballots were cast in the 2020 election. The Census Bureau’s report found 159.9 million ballots were cast in the election.

“The Census Bureau’s library of administrative records makes it uniquely capable of compiling and publishing statistics on election crime and other forms of fraud,” the report concludes, echoing the Trump administration’s prior attempts to link election and social service spending fraud to immigration.

The Census Bureau report comes after Trump has sought to lean on the statistical agency to uphold his administration’s monomaniacal goal of delegitimizing the results of the 2020 presidential election. Trump used the report’s findings to urge the passing of the SAVE America Act, a sweeping voter suppression bill that Trump has been pressuring Congress to pass for months, but which does not have widespread support even among Republicans  in Congress. 

“On the first 128,000,000 Voters,” Trump wrote on Truth Social, “the Census Bureau proves that over 24,000 non-citizens voted illegally! The Census Bureau is going to analyze the next 32,000,000 Voters, and this number will explode. I WON THE ELECTION!”

Howard Lutnick, the secretary of the Department of Commerce, under which the Census Bureau falls, backed his boss in a post on X touting the report.

Experts have told TPM that there is “absolutely no basis” to Trump and the MAGA movement’s claims about noncitizen voting. Trump and Republicans have been propping up conspiracy theories about noncitizens voting en masse for Democrats in federal elections for years. Nevermind that the findings of Tuesday’s Census Bureau report obviously does not support any claim that the 2020 election was somehow rigged by a crush of ineligible, noncitizen voters. The bureau’s analysis diverts from normal operations in several ways, researchers and public data experts said Tuesday, which makes the veracity of its findings impossible to confirm.

“It lacks transparency and accountability,” Beth Jarosz, a senior fellow in demographic and socioeconomic analysis at the Georgetown University Massive Data Institute, told TPM.

Neither the Census Bureau nor the Commerce Department immediately responded to TPM requests for comment.

The Census Bureau used a roll of private voter data, compared with various federal data sources, to substantiate its finding. That’s unprecedented, Jarosz said.

“[O]ne of the red flags in the report is that the Census Bureau, when they start to use a new data source, typically goes through a very extensive process of trying to understand the data quality,” Jarosz said. “And I don’t see any evidence that they did that.”

On X, Omar Noureldin, senior vice president of policy and litigation strategy at Common Cause, said the Census Bureau doesn’t collect the information it would have needed to publish definitive findings about alleged noncitizen voting.

“The 2020 Census did not collect citizenship status, and the Census Bureau does not have a comprehensive individual ‘citizenship records’ file to definitively match against voter rolls,” Noureldin wrote. The type of data the Bureau did use in its report, according to its own methodology, “are modeled population estimates, not evidence that 24,000 people voted illegally,” he continued. 

The report doesn’t list any authors or even a bureau department from which the report came. According to Mike Schneider, an Associated Press reporter who covers the Census Bureau, that’s atypical.

“Every other report on the bureau’s ‘Census Publications’ site is credited to a statistician, economist, demographer or researcher,” Schneider wrote on his Substack.

Not even a department within the bureau is listed on the completely anonymous report. The report says its data was analyzed and published with “high confidence,” but lacks any definition on what high confidence means.

Tuesday’s report references legislation from 1902 which made the Census Bureau into a permanent governmental entity and empowers the Bureau to “collect statistics related to… crime, pauperism, and benevolence, including prisoners, paupers, juvenile delinquents,” and the like. The report’s authors cite this provision as giving the Bureau authority to “compile and

publish information on crime.”

This is an oversimplified interpretation of that provision, Mark Mather, the associate vice president of programs at the Population Reference Bureau, said in a data user forum. The legislation’s provision authorized the decennial Census to count “people already in institutions,” Mather wrote, not to identify “potential lawbreakers.”

“I’m concerned about the potential chilling effect of this work,” Mather continued, saying the report “sends a potentially troubling message about how administrative data shared with the Census Bureau might be used.”

Just in the last two months, the administration has:

• Thrown out long-awaited plans to update race and ethnicity data collection;
• Eliminated an umbrella of highly technical, statistical privacy methods, which will result in less published data from the bureau;
• Proposed adding a question to the 2030 Census to identify and eliminate noncitizens from the total population count, and;
• Proposed a regulation that would eliminate the collection of race and ethnicity data for the first time in census history.

The Department of Justice since 2025 has continued escalating its efforts to obtain voter roll data from states. Federal judges have dismissed 23 DOJ suits against states which have refused to turn over voter information.