fb tracking

Public Citizen Testimony in Support of SB 893 — Prohibiting AI Deepfakes in Elections

Public Citizen Testimony in Support of SB 893 -- Prohibiting AI Deepfakes in Elections

To: Chairman Bryan Hughes and the Members of the Senate Committee on State Affairs
CC: Sen. Angela Paxton, Sen. Paul Bettencourt, Sen. Brian Birdwell, Sen. Bob Hall, Sen. Adam Hinojosa, Sen. Mayes Middleton, Sen. Tan Parker, Sen. Charles Perry, Sen. Charles Schwertner, Sen. Judith Zaffirini 

Via hand delivery and by email. 

From: Adrian Shelley, Public Citizen, ashelley@citizen.org, 512-477-1155 

Re: SB 893, AI deepfakes in elections – Public Citizen testimony in support 

Dear Chairman Hughes and Members of the Committee: 

Public Citizen appreciates the opportunity to testify in support of SB 893, relating to criminal offenses for creating and distributing certain misleading images and videos; creating a criminal offense. We support this bill because it broadens the law criminalizing the use of deepfakes to include images, because it adds an intent requirement to the offense, and because it removes the 30-day timing requirement. 

AI-generated deepfakes are a here-and-now threat to our elections. “Deepfakes” are fabricated content depicting a candidate saying or doing things they never did—deliberately designed to damage that candidate’s reputation or to deceive voters.  

Recent advances in the realm of AI have made it such that tools needed to create deepfakes are now widely accessible. Meanwhile, the quality of deepfake technology is improving rapidly, making it harder for the average person to detect a deepfake. 

Take for example this Desantis Ad which used a deepfake to depict Donald Trump hugging Dr. Fauci. Or this damaging deepfake audio of a party leader in Slovakia circulated two days before the election. 

Without regulation, deepfakes are likely to further exacerbate voter confusion and a loss of confidence in elections. A deepfake video could be released shortly before an election, with no time to determine its fraudulency – ultimately misleading voters and changing the outcome of the election. 

An intent to deceive requirement is added to existing law. 

SB 893 adds an intent requirement to the existing law banning deepfake videos in elections. A person only commits an offense if they create and distribute a deepfake video or image with the intent to deceive and cause injury. This intent requirement ensures that only malicious actors will be punished under the deepfake ban. 

SB 893 also changes the definition of deepfake video to remove the intent requirement from the definition. This new definition acknowledged that a video created with artificial intelligence is still a deepfake absent the intent requirement of the creator. By putting an intent requirement in the offense itself, the bill will limit the law to punishing only those with malicious intent. 

The bill also adds an affirmative defense to the law if a deepfake image includes a clear label stating that the image is altered. A clear label would help viewers to identify a deepfake image and make clear that its creator did not intend to deceive viewers.  

Images are added to the law banning deepfakes. 

SB 893 also creates a new crime of creating a deepfake image with the intent to deceive. The existing state law only applied to videos. The creation of a video is a Class A misdemeanor and the creation of an image is a lesser offense of a Class B misdemeanor. The section adding images excludes cartoons, caricatures and other clearly satirical images, as well as the use of conventional image editing techniques such as adjusting saturation, brightness, and color. 

The 30-days before an election requirement is removed from the law. 

Finally, this bill removes from the offense a requirement that a deepfake video be published or distributed within 30 days of an election. This timing requirement was arbitrary—a deepfake video can have a negative effect on an election well before 30 days. Furthermore, because an intent requirement is added to the law, the timing of the publication becomes irrelevant. It may be that the original inclusion of a 30-day requirement was meant to function like an intent requirement—with the proximity to the election taken to indicate malicious intent. A clean intent requirement in the law makes this timing element obsolete. 

In conclusion, we ask you to support SB 893 because we believe the bill will strengthen the law banning deepfakes and discourage their use to improperly influence elections. It adds an intent requirement to the deepfake law, adds deepfake images, and removes the timing requirement.