Joe Biden-Robert Hur interview audio wasn’t leaked; it’s a deepfake, DOJ and experts say

Republicans in the U.S. House of Representatives have threatened to hold Attorney General Merrick Garland in contempt for refusing to release an audio recording of Special Counsel Robert Hur’s interview of President Joe Biden in a classified documents case.

A June 5 TikTok video claimed an audio clip from that interview may have already leaked.

"Is this a leaked audio of special counsel Hur questioning Joe Biden," the video’s caption said. "Is it real? If it's fake, Joe Biden must release the real one. Otherwise, how can it be disputed?" 

We found other social media posts sharing the same video.

TikTok identified this video as part of its efforts to counter inauthentic, misleading or false content. (Read more about PolitiFact’s partnership with TikTok.)

The Justice Department has been loath to release the audio recording of the Biden-Hur interview, partly, it says, because of the potential the audio could be manipulated using artificial intelligence to portray Biden or Hur saying something they did not say.

The concern about deepfake audio recordings is real. As AI-generated audio recordings increase, the tools that can detect them lag, as PolitiFact reported in March. In January, a fake robocall mimicking Biden’s voice urged Democrats not to vote in New Hampshire primaries. Steven Kramer, a political consultant, was charged with numerous crimes  in that case related to voter suppression and impersonating a candidate.

A Justice Department spokesperson said the department was confident the recording is fake, and deepfake experts told PolitiFact the recording had signs of manipulation. We also searched Google and the Nexis news database and found no reports of audio from the Biden-Hur interview being leaked. 

(TikTok screenshot)

"Although this is a pretty good deepfake of Biden’s voice, there is still a cadence in his voice that has the tell-tale signs of AI generation," said Hany Farid, a University of California, Berkeley, computer science professor and digital forensics expert.

The voices of Biden and Hur may sound comparable to their real voices and use similar language to the official transcript released by the Justice Department, but the audio in the TikTok video differs significantly with the transcript.

Also, a transcript shown on-screen in the TikTok video also doesn’t match the released Biden-Hur interview transcript, omitting some words and adding others.

Here’s what the TikTok video says the transcript and audio say:

Mr. Hur: I wanted to ask some questions related to duties performed by — I guess I’ll call them front-office staff. Is that an accurate description?

President Biden: Ah, I don’t really know, but OK.

Mr. Hur: OK. Well what did you call the people who worked for you?

President Biden: Uhh, by their names if I knew them.

Mr. Hur: OK.

President Biden: I’m not being facetious.

Mr. Hur: No of course not.

President Biden: I’m trying to think back ten years now, but — or more than 10 years, I guess. No, about ten years. No, wait. 2016, 2016, 2017, 2018, five, six, anyway.

Mr. Hur: Mr. President, let’s continue.

Here’s what the official transcript says:

The official transcript of the interview contains some words in the video, but is significantly different. 

Mr. Hur: We've gotten to know the names of some of the folks who worked in your front-office staff during your vice presidency, (name redacted), Michele Smith, (name redacted) and (name redacted) who then became (name redacted). So I wanted to ask some questions relating to duties that those — I guess I‘ll call them front-office staff. Is that an accurate description of them? 

President Biden:  I never did, but sure. 

Mr. Hur: OK. Well, what did you call them? 

President Biden: By their first names. 

Mr. Hur: OK.

President Biden: I‘m not being facetious. 

Mr. Hur: No. 

President Biden: I'm not being facetious. And, by the way, a lot of this stuff would come back and whoever — if my intelligence team came in, and they gave me something, they, they’d come and pick it up, whatever it was. 

So, for example, now, I’m trying to think back 10 years, but — or more than 10 years, I guess. No, about 10 years. But today, I had a briefing with — about what's going on in Israel, a detailed briefing, in my residence upstairs in a room that’s not dissimilar to this off of the bedroom.

A Justice Department spokesperson told PolitiFact in an email that the actual audio matches the actual transcript and is confident the TikTok video is fake. 

Featured Fact-check

Barack Obama

stated on January 31, 2014 in an interview with CNN

"What is and isn't a Schedule I narcotic is a job for Congress."

The video is an example of why the department argued in court that not releasing the audio would make identifying fakes easier, the spokesperson said.

In a lawsuit brought by a coalition of conservative groups, including Judicial Watch and the Heritage Foundation, and media outlets including The Associated Press and CNN, the Justice Department expressed concerns in a May 31 court filing that the Biden-Hur audio recording would be used to create deepfakes.

"​The passage of time and advancements in audio, artificial intelligence, and ‘deep fake’ technologies only amplify concerns about malicious manipulation of audio files," the department wrote. "If the audio recording is released here, it is easy to foresee that it could be improperly altered, and that the altered file could be passed off as an authentic recording and widely distributed." 

Bradley Weinsheimer, a Justice Department associate deputy attorney general, wrote in an affidavit that "malicious actors" could manipulate the recording. He said if the recording was not released, the department would be "better able to establish the illegitimacy of any malicious deepfake."

Weinsheimer said the written transcript accurately portrays the words in the audio recording, except for "minor instances" such as the use of filler words, including "um or uh," or words that were repeated, such as "and, and." Aside from that, he wrote, there are "no material differences between the audio recording and transcripts."

Weinsheimer said Hur and other FBI personnel who attended the interview confirmed the transcript accurately portrays the interview.

In May, the Republican-led House Judiciary and Oversight committees voted to advance contempt proceedings against Garland for his refusal to turn over audio records of Hur’s interview of Biden.

Garland testified June 4 before the House Judiciary Committee and said that the Justice Department had gone to "extraordinary lengths" to respond to Congress’ "legitimate request" for information, including providing legislators with Hur’s report, allowing Hur to testify before Congress for more than five hours and providing a transcript of the interview.

"But we have made clear that we will not provide audio recordings from which the transcripts that you already have were created. Releasing the audio would chill cooperation with the department in future investigations," Garland said. "And it could influence witnesses’ answers if they thought the audio of their law enforcement interviews would be broadcast to Congress and the public."

Experts told PolitiFact the audio has the markings of a deepfake.

"The most relevant indicators reflect inconsistencies in speech rhythm and pitch, unnatural intonations, and lack of alignment between content in the audio" and the real transcript, said Manjeet Rege, director of the Center for Applied Artificial Intelligence at the University of St. Thomas. "The rhythm and cadence don't align with what we know of President Biden’s normal speaking style."

He said technology has a hard time replicating the details of pitch, tone and pace.

Hafiz Malik, a University of Michigan-Dearborn electrical and computer engineering professor, researches deepfakes using a software he created with his team that compares potential deepfake video and audio with hundreds of hours of authentic audio of public figures, including Biden.

"Based on my analysis, it is a deepfake," Malik told PolitiFact.

Malik said in machine-generated content, he closely examines the pitch of the speaker. In normal conversation, speakers' emotions and thought processes change — and their pitch also changes with them.

"It doesn't stay constant. It is changing every moment," Malik said, adding that Biden's voice in the TikTok recording was monotonous.

He also looked at high-frequency bands from the audio. Those fade away in real voice, he said, but not in machine-generated content. Finally, he said in natural speech, phrases differ in length, but in machine-generated speech such as this, they are shorter.

A TikTok video that claims it may be a leaked clip of an audio recording between Biden and Hur includes audio and a transcript that differs from the official transcript released by the Justice Department. 

Hur, according to a Justice Department official, confirmed that the official transcript accurately reflects the audio interview, which the department has declined so far to release.

There are no credible news reports about a portion of the audio recording being leaked. The Justice Department and three deepfake experts said the audio is a deepfake.

The claim is False.

PolitiFact Researcher Caryn Baird contributed to this fact-check.