The Deepfake Watchlist is Resemble AI's weekly surveillance of synthetic media incidents, ongoing cases, and disputed content shaping the news cycle. Each week we track confirmed incidents, emerging attack vectors, and claims under investigation, alongside the provenance, detection, and policy threads running underneath them. New to the world of deepfakes? Read the Deepfake 101 Guide to learn more about how to protect yourself and your company from threats.
1. Mark Cuban calls out Meta for running AI-generated scam ads using his likeness
Speaking on the August 21 episode of Pivot with Kara Swisher, Mark Cuban said Meta continues to allow AI-generated ads featuring his image and voice to promote products he has never endorsed, including fake diabetes cures sold to elderly users for hundreds of dollars (Parade, August 24).
- Category: Brand / Likeness
- Type: Attack
- Modality: Image, Video
- Policy / Regulatory: Meta has removed more than 340,000 nudify ads and faced bipartisan congressional pressure over deepfake advertising, but no regulatory mechanism currently requires pre-publication verification of celebrity likenesses in paid ads.
- Trend: AI-generated celebrity endorsement scams are a recurring, unresolved platform moderation failure, with the same vector producing documented harm week after week despite public commitments to enforcement.
- Attack vector: AI-generated likeness of a public figure used in paid social media advertisements to promote fraudulent health products, targeting elderly users through Meta's ad platform.
Cuban said he receives messages every week from elderly people who believe they purchased a product he endorsed. "Every single week I get emails from little old men and ladies saying, 'Will my cure for diabetes work?'" he said. He called Facebook "the worst of the worst" and said the company knows the ads are fraudulent.
Issue 14 of this Watchlist covered Meta running 32 ads for the Kromix nudify app, from a brand-new account with zero followers, for up to 46 hours before a journalist flagged them. Cuban's complaint lands on the same structural problem: Meta's ad review system is designed for point-in-time approval, and AI-generated impersonation ads keep getting through because the review happens before the content runs, not while it is running.
2. Congressional candidate used AI to fabricate endorsement from deceased Hawaiian activist
Hawaii News Now's AI-generated endorsement depicting late activist outrages Hawaiian community reports that congressional candidate Nathan Berning used artificial intelligence to create a 30-second campaign ad featuring the voice and image of the late Native Hawaiian activist Haunani-Kay Trask, who died in 2021, appearing to endorse his candidacy.
- Category: Political / Electoral
- Type: Attack
- Modality: Image, Audio
- Policy / Regulatory: Hawaii's Act 247, which took effect this year and prohibits harmful unauthorized AI-generated likenesses, may give the Trask family a civil cause of action; the family has already filed a cease and desist.
- Trend: AI-generated campaign content is escalating from attack ads depicting living opponents to fabricated endorsements from deceased public figures.
- Attack vector: AI-generated voice and image of a deceased public figure used in a television campaign ad to fabricate a political endorsement that never existed.
Berning initially told Hawaii Public Radio he had never heard of Trask and that the woman in the ad was a model. That was disproven because the AI-generated photo was a widely published image of Trask. He later admitted the ad was AI-generated. Trask's sister, Mililani Trask, called it a violation and a perpetration of fraud on voters.
The ad is part of a coordinated tactic funded by anti-abortion activist Randall Terry's PAC, which backs at least 10 similar candidates nationally. Federal rules prohibiting censorship of congressional ads mean stations cannot refuse to run the content. Hawaii State Rep. Scot Matayoshi pointed to Act 247 as a potential remedy, calling it exactly the scenario the legislature anticipated.
3. Houston man arrested for thousands of AI-generated child sexual abuse images
KHOU's Humble man accused of possessing thousands of AI-generated child pornography images reports that Harris County Precinct 1 Constable Alan Rosen announced the arrest of Alexander Bigelow, 29, after detectives found thousands of AI-generated child sexual abuse images on a computer and hard drive at his Humble, Texas home.
- Category: CSAM / NCII
- Type: Response
- Modality: Image
- Policy / Regulatory: Bigelow faces three second-degree felonies under Texas law for possession of computer-generated pornography of children under ten, each carrying up to 20 years in prison.
- Trend: Texas is emerging as an early proving ground for AI-generated CSAM prosecution under its SB 20 framework, with multiple arrests in 2026.
- Attack vector: AI image generation tools used to produce thousands of child sexual abuse images stored locally on personal hardware.
This is the third major AI-generated CSAM arrest in Texas in 2026. In January, Kane James Kellum was indicted after police found he had used AI tools to generate explicit imagery of children he personally knew. In May, Adan Covarrubias was arrested after allegedly scraping photos from South Texas school social media pages and using AI tools to create imagery of approximately 30 identified victims.
All three cases produced charges under Texas Senate Bill 20, which went into effect in September 2025 and explicitly criminalized possession of computer-generated child pornography regardless of whether the imagery depicts a real child. Three prosecutions within eight months suggests the statute is functioning as intended.
4. Gay dating app Goose exposed for using AI-generated influencers to lure users
Futurism's Dating App Cofounder Bails After Scandal Over Using Fake AI-Generated Influencers reports that Derek Chadwick, cofounder of the gay dating app Goose, resigned after the platform was exposed for using AI-generated influencer accounts on Instagram to send targeted direct messages recruiting users to join the app.
- Category: Fraud / Impersonation
- Type: Attack
- Modality: Image
- Policy / Regulatory: No regulatory action has been taken; the use of fabricated identities to drive signups is a potential consumer protection issue.
- Trend: AI-generated personas are moving beyond scam operations and into legitimate product marketing, blurring the line between synthetic media fraud and deceptive advertising.
- Attack vector: AI-generated influencer profiles on Instagram used to send direct messages to potential users, impersonating real community members to manufacture trust and drive app downloads.
WIRED first reported in July that Instagram accounts promoting Goose appeared to be AI-generated, with fabricated identities sending messages accompanied by invite codes. Chadwick initially denied the allegations, then resigned weeks later, writing that his vision and the company's direction were no longer aligned.
The mechanism is the same one that powers romance scams and catfishing: a synthetic identity designed to make someone believe they are interacting with a real person. The Guardian recently estimated that 40 to 60 percent of branded influencer content involves undisclosed AI-generated accounts, suggesting Goose's tactic may be more common than the backlash implies.
5. Montana deepfake disclosure law faces First Amendment challenge in federal court
Missoula Current's Political complaints by MT legislators over AI dropped and KTVH's Montana law restricting AI-generated political content heard in federal court report that U.S. District Judge Susan Waters heard initial arguments on August 21 in a challenge to Montana's Senate Bill 25, which requires disclosure when AI-generated deepfake content is used in campaign materials within 60 days of an election.
- Category: Political / Electoral
- Type: Response
- Modality: Image
- Policy / Regulatory: The lawsuit challenges SB 25 on First Amendment and due process grounds; the Commissioner of Political Practices separately dismissed three complaints, finding the AI-generated mailers fell under a satire exemption.
- Trend: State deepfake disclosure laws are facing their first constitutional stress tests, with outcomes that could determine whether the regulatory approach of 29 states survives judicial scrutiny.
- Attack vector: AI-generated political mailers depicting state legislative candidates with Pride flags and pronoun buttons, distributed by a conservative PAC without AI disclosure labels.
The case originated when a PAC sent hundreds of AI-generated mailers targeting three Republican candidates. The PAC's attorneys argue SB 25 forces political speakers to label their own protected speech as false, calling it unconstitutional compelled speech. The Commissioner dismissed the complaints on satire grounds.
This week, Axios reported that Massachusetts has passed deepfake legislation through the House but the Senate has let both bills sit, leaving voters unprotected heading into the midterms, even as an AI-cloned version of Governor Maura Healey's voice appeared in a Republican rival's ad. Judge Waters indicated she will rule by early September, and an adverse decision could destabilize the disclosure-based framework that most of the 29 states with election deepfake laws have adopted.
Honorable mentions
7NEWS Australia reported that an AI voice cloning website with more than two million voice models has scraped the voices of Australian celebrities, the Prime Minister, and everyday citizens, needing just 10 seconds of audio to produce a clone.
Singapore's police force issued new codes of practice on August 18 requiring Facebook, Instagram, and TikTok to verify advertiser identities against government records and block ads from unlicensed financial firms, with fines of up to S$1 million.
The pattern
.jpg)
- The dead are becoming the newest attack surface. Carl Sagan's voice used to sell an AI product without his estate's consent. Haunani-Kay Trask's face and voice fabricated into a political endorsement five years after her death. Both exploit the same gap: the people being impersonated cannot object, and the legal frameworks that protect living individuals are only beginning to extend to the deceased.
- The enforcement infrastructure for AI-generated CSAM is producing results. Three AI-generated CSAM cases in Texas in 2026, all producing charges under SB 20, suggest that criminalizing computer-generated child exploitation material regardless of whether a real child is depicted is translating into actual prosecution. The question is whether other states follow Texas's lead or wait for federal legislation.
- AI-generated deception is normalizing faster than disclosure norms can keep pace. Goose used fabricated influencers to grow a consumer product. A Montana PAC sent AI-generated mailers. A congressional candidate used a dead activist's likeness and initially denied knowing who she was. In each case, the creator's first response was denial rather than disclosure. The Montana court challenge is the legal expression of this tension: whether mandatory labeling survives First Amendment scrutiny will determine whether the regulatory response has any teeth.
Watching next week
- Montana SB 25 ruling timeline. Judge Waters indicated a decision by early September; an adverse ruling could destabilize the disclosure-based framework that 29 states have adopted.
- Carl Sagan / Luma AI proceedings. Whether Luma responds to the complaint or seeks to settle will signal how generative AI companies handle direct likeness infringement claims.
- Texas AI-generated CSAM prosecution pipeline. Three cases producing charges in 2026; the pace of new investigations will indicate whether SB 20 is building systemic deterrence.
- Massachusetts Senate action on deepfake bills. Whether the Senate moves before the election or lets them die will determine whether Massachusetts voters have any AI election protections for the midterms.
The Deepfake Watchlist publishes every Thursday. Subscribe to receive it in your inbox, or follow Zohaib Ahmed on LinkedIn for the weekly social companion. Track every documented incident in the Resemble Deepfake Incident Database, and read the full methodology in our 2026 Midyear Deepfake Threat Report.