Republicans Posted a Deepfake of a Senate Candidate. It Was Technically Legal.

On March 11, 2026, the National Republican Senatorial Committee posted a video of James Talarico to X.

Talarico is the Democratic nominee for Senate in Texas. In the 85-second video, he appears to look directly at the camera, speaking in his own voice, reading excerpts from his past tweets on transgender issues, immigration, Christianity, and Planned Parenthood. He looks like Talarico. He sounds like Talarico.

He wasn't Talarico. The person in the video was an AI-generated model of the candidate.

There was a disclosure in the corner of the screen. Four words: "AI GENERATED." Small. Faint. Confined to the bottom-right corner.

The NRSC did not comment further on the ad. A source familiar with their thinking described AI as "a consistently effective way to highlight opposing candidates' statements" and noted that the words in the video were drawn from Talarico's real posts.

---

The Law That Didn't Stop It

Texas was the first US state to ban deepfakes in political campaign ads. It passed that law years ago.

The NRSC's video ran in Texas anyway.

Critics, including advocacy group Public Citizen, say the law has critical gaps. The disclosure label — whatever its size or legibility — may satisfy the technical requirement. There's no minimum size standard for the "AI GENERATED" text. There's no requirement that viewers actually notice it. The law bans deceptive deepfakes; an ad that technically discloses its nature may fall outside that category even if the disclosure is nearly invisible in practice.

CNN's review of the video found the fake Talarico "looks uncannily like the actual candidate." The blazer, the shirt, the voice, the cadence. A viewer who missed the four-word corner label would have no reason to believe they weren't watching the real person.

The NRSC saw no downside to running it.

---

Not the First, Not the Last

This wasn't a bolt from the blue. In October 2025, the same NRSC ran an attack ad featuring a deepfake of Senator Chuck Schumer. That ad also ran in a state with deepfake disclosure requirements.

Multiple other campaigns have used AI this cycle. An attack ad against Texas congressional candidate Wesley Hunt showed him holding a Pomeranian in fake scenes to depict Hunt as a "show dog." That ad included no AI disclosure at all.

California Governor Gavin Newsom posted AI-generated videos of Trump and administration officials "crying in handcuffs." Those were clearly satirical; no one would mistake them for real. The Talarico video was not clearly satirical.

CNN described "phony videos proliferating in midterm races." Roughly half of US states have passed some form of deepfake election law. Most require only disclosure. None have been tested at scale against a campaign prepared to push legal limits.

Zero federal laws govern the practice.

---

The Playbook Is Now Established

What the Talarico ad confirmed is a playbook: take a real candidate's real words, generate a photorealistic AI version of that candidate delivering those words in the most damaging possible framing, add a small disclosure label, and post it.

You're not fabricating statements. You're not putting words in anyone's mouth that they didn't say in some form. You're generating a realistic visual wrapper around real content. The disclosure, however faint, provides legal cover.

The NRSC source called it "consistently effective." That's not a moral judgment. It's an operational one. The ad reached people. It was shared. Talarico's campaign pushed back, with communications director JT Ennis accusing rivals of "misleading Texans." The campaign's response amplified the video's reach.

This is how it works now. The question isn't whether a political party's deepfake of a Senate candidate is crossing a line. It's whether the line will be moved before November.

---

What November Looks Like

The 2026 midterms are eight months away. The Senate map has several competitive races. Texas is one of them — Talarico is running against an incumbent in a state Republicans can't afford to lose.

There's no federal legislation moving fast enough to matter. The EU's Article 50 deepfake labelling law takes effect August 2 — three months before US midterms — but it covers EU platforms and EU elections. It has no jurisdiction over X, Facebook, or Texas Senate races.

Public Citizen has called on the FEC to treat photorealistic AI depictions of candidates as fraudulent misrepresentation. The FEC has not moved.

What's already established is the precedent. A national party committee used a photorealistic AI-generated video of an opposition Senate candidate, ran it through legal review, concluded that a faint corner label provided sufficient disclosure, and posted it. They called it effective. They have the same tools available for every competitive race between now and November.

The candidates who get deepfaked next may not be as quick to respond. The voters watching those videos may not notice the label in the corner.

The information environment for the 2026 midterms isn't what it was in 2022. The tools changed faster than the rules.

For Albis's full tracking of how AI is reshaping information integrity, see the Trust Collapse cascade tracker and the GAI Index monitoring blind spots in AI political coverage.