The Deepfake Watchlist: Week of October 23–29, 2026

‍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.

★ Featured: A Texas campaign posted AI-generated Black voters thanking the candidate for jobs, food, and help with the light bill

Newsweek's Texas Republican judge faces backlash over AI endorsement video reports that interim Fort Bend County Judge Daniel Wong's campaign posted an AI-generated video on October 23 showing Black residents endorsing him, then removed it after viewers flagged it, with his communications director confirming to FOX 26 Houston that the video is AI-generated.

• Category: Political / Electoral
• Type: Attack
• Modality: Video, Audio
• Policy / Regulatory: No complaint has been filed, and Texas's election deepfake law is written around fabricated depictions meant to injure a candidate, so whether it reaches fabricated supporters is untested.
• Trend: Campaigns using synthetic media to manufacture the appearance of community support, rather than to fabricate statements by an opponent.
• Attack vector: AI-generated testimonial video posted to an official campaign Facebook page and reshared by the county party during early voting, eleven days before the November 3 election.
• What we saw in the content: Reporting on the video documented the kinds of signals detection models are trained to flag, for example:
  ‍
  • A speaker wearing a "Daniel Wong" hat with the candidate's name misspelled, a common failure when generation models render text
  • A roadside marquee behind one speaker covered in illegible letters, another text-rendering artifact
  • Voices that viewers noticed sounded alike across speakers who were presented as different people
  • Visual distortions that prompted viewers to question the video shortly after it was posted

Most of the election deepfakes we have tracked this cycle put fabricated words in an opponent's mouth, and the Fort Bend video went after something different: the appearance of support itself. The people in it thank Wong for jobs, groceries, and keeping the lights on, which is why his opponent, Commissioner Dexter McCoy, called it reprehensible for reducing Black voters to people looking for a handout. The Wong campaign says the video was made by a supporter who is a Black Democrat and that the creator told them the people depicted gave consent, a claim Newsweek says it has not independently verified.

The forensic tells here were not subtle, and viewers caught them quickly. A misspelled name on a hat and a marquee full of nonsense letters are the kinds of text-rendering failures that still show up in consumer generation tools, and similar-sounding voices across supposedly different speakers are a classic sign of a single voice model doing all the talking. Even so, the video went out from an official campaign page and the county Republican Party's page during early voting, and for a voter scrolling past it, where it appeared carries more weight than who originally made it.

Election deepfake laws, including Texas's, were written for a world where the deceptive content depicts a candidate, and a fabricated crowd of supporters falls awkwardly between categories. Provenance at the point of generation would have labeled this video as synthetic no matter who made it or which page posted it, which is a protection that does not depend on lawmakers anticipating every new way campaigns find to use these tools.

1. UK police: £24.3 million lost to AI-enabled investment fraud in eight months, mostly by people over 60

The City of London Police's Report Fraud warning after £24.3 million stolen through AI enabled investment endorsement schemes reports that between January and August 2026, Report Fraud received more than 1,300 reports of investment fraud where AI, including deepfakes, was used to make the offer more convincing, with people aged 60 or over making nearly 80 percent of those reports.

• Category: Fraud / Impersonation
• Type: Attack
• Modality: Video, Image
• Policy / Regulatory: Report Fraud reviewed each report individually to confirm AI involvement, and police are directing would-be investors to the FCA's Firm Checker before transferring money.
• Trend: Synthetic celebrity and expert endorsements becoming a routine opener for investment fraud aimed at older adults.
• Attack vector: AI-generated videos and images of trusted public figures endorsing fake investment schemes, used to build credibility before money changes hands.

Synthetic endorsement fraud is landing hardest on people over 60, the same profile as the 88-year-old British Columbia man we covered last week who lost $450,000 after seeing a deepfake of Prime Minister Mark Carney. This is worrisome for all of us with parents and grandparents who trust a familiar face on a screen.

It also matters where these endorsements live. Our own scan of Meta's Ad Library, published last week, found crypto and brokerage brands running AI-flagged creative at roughly three times the study's overall rate, and pages using a brand's name without being run by the brand at more than ten times the rate of the brands themselves. The ad system is the front door for much of this fraud, and checking a firm's registration only helps once someone already suspects the endorsement.

2. Israel's election ends inconclusively after a campaign fought with deepfakes from both camps

Al Jazeera's Bots, deepfakes and surveillance: Israel's election goes to war online reports that both Prime Minister Benjamin Netanyahu's camp and the opposition led by Gadi Eisenkot pushed AI-generated and deepfake videos on X ahead of the October 27 vote.

• On October 25, Netanyahu shared an AI-generated image of a nuclear explosion over Tel Aviv, according to The Jerusalem Post, prompting opposition leaders to accuse him of trying to frighten voters.
• On polling day, Netanyahu posted photos he said showed crowds lining up to vote in Arab-majority communities, and at least one was labeled AI-generated after BBC analysis, ABC News Australia reports.
• Category: Political / Electoral
• Type: Attack
• Modality: Video, Image
• Policy / Regulatory: Israel's Central Elections Committee had already restricted one Likud turnout app over privacy concerns, but the reporting describes no rule that kept campaigns or a sitting prime minister from posting synthetic imagery during the campaign or on election day.
• Trend: Heads of government and major parties posting AI-generated imagery from official accounts during national elections.
• Attack vector: AI-generated video and imagery distributed through candidates' own verified social media accounts, amplified by bot networks.

The nuclear image is a hypothetical rendered vividly, and most viewers would understand it as an illustration of a threat rather than a photograph of an event. The polling-day image is different and more concerning, because it was presented as documentation of something happening that day, at the moment voters were deciding whether to turn out.

When synthetic media comes from the official account of a head of government, the usual defenses lose their footing. Platform policies and fact-checkers are built to flag anonymous fakes, and a verified account carries a presumption of authenticity that a fabricated image can borrow. With the result now disputed and coalition talks underway, Israel is a preview of what an election looks like when every side treats synthetic media as ordinary campaign material.

3. Queensland moves to criminalize the creation of sexual deepfakes, not just their distribution

ABC News's Queensland government introduces new laws criminalising creation of sexually explicit deepfake images reports that Attorney-General Deb Frecklington introduced legislation on October 27 making it a crime to create non-consensual, sexually explicit deepfake images or audio, with maximum penalties of three years in jail for adults and 18 months for children.

• Category: CSAM / NCII
• Type: Response
• Modality: Image, Audio
• Policy / Regulatory: Queensland currently prohibits distribution but not creation, and the bill also adds a definition of intimate audio material and expands "intimate image" to cover non-genuine images.
• Trend: Lawmakers moving upstream from distribution to the act of generation itself.
• Attack vector: Not applicable, this is a legislative response.

The case that exposed the gap was Antonio Rotondo, who created sexualized deepfakes of current and former students and staff at a Brisbane school, including a 13-year-old, and could only be punished for distributing them. Frecklington said schools are on the frontline, with fake images being used to bully both students and teachers, and the mother of a girl who was bullied with fake nude imagery has publicly backed the change.

The definitional work in the bill deserves as much attention as the headline offence. Covering "non-genuine" images and intimate audio closes the arguments offenders have used to say an image does not count because the setting is impossible or part of the body is obscured. Making creation itself the crime matters because by the time an image is distributed, the harm to the person depicted is already very hard to undo.

4. An attacker posed as 32 different AI agents to hunt for passwords and cloud keys

Dark Reading's Attacker Relentlessly Scans for Credentials by Impersonating AI Bots reports that a single threat actor rotated through about 32 fake AI agent identities, including ChatGPT, Claude, and Perplexity, to scan organizations in financial services, retail, telecom, energy, and software for exposed passwords, API keys, and cloud credentials from late September through late October.

• Category: Fraud / Impersonation
• Type: Attack
• Modality: Agentic AI
• Policy / Regulatory: No common standard exists for verifying that a self-identified AI agent is who it claims to be, and several major US banks have restricted third-party agent access to customer accounts this month.
• Trend: Attackers impersonating trusted AI agents to slip past the allow-lists companies are building as legitimate agent traffic grows.
• Attack vector: An automated scanner on rotating cloud addresses that presents spoofed AI agent identities while requesting files that commonly store secrets, such as configuration files, cloud credential files, and private SSH keys.

Most of what we track in the Watchlist is impersonation of some kind, and this campaign points the same tactic at a newer kind of identity. Researchers found that on one day in October the scanner reached 48 percent of the monitored organizations that saw any AI bot traffic, and the same machine claimed to be an OpenAI, Anthropic, Meta, and Perplexity agent within seconds. Its fake Claude identities all carried the contact address of Anthropic's real crawler, a copied string that shows how little effort the disguise took.

The disguise works because most systems still decide who a visitor is by what the visitor says about itself. At several organizations the impersonator made up more than 80 percent of the traffic labeled as AI bot traffic, which means the numbers companies use to set agent policy can be badly skewed. That is why our own Agent Detection reads how a session behaves rather than the identity it presents, since a claimed name is the easiest part of an agent to fake. Last week, Axios reported that bank executives struggle to tell an authorized agent from a rogue one, and that question gets harder once attackers start wearing the names of the agents banks have chosen to trust.

Honorable mentions

Mississippi AG forces removal of Scott Colom's AI ad depicting Cindy Hyde-Smith. Attorney General Lynn Fitch said on October 23 that the Democratic Senate nominee likely broke state law with an undisclosed AI-generated ad showing his Republican opponent running from reporters, and Colom deleted it rather than face litigation, Mississippi Today reports. Mississippi's law exempts satire and parody, and Colom maintains the ad portrays a true story. The Mississippi Free Press has also reported on Hyde-Smith's campaign building a fake website using Colom's name and an old mugshot.

Nissan replaces AI-edited press photos of its SEMA concept car. Motor1 and The Autopian flagged anomalies in Nissan's official images of the Sentra Spec-V Concept, and Nissan confirmed an external creative partner had used AI-assisted editing tools on the backgrounds, introducing errors into vehicle details. Nissan is replacing the images, and the car itself is real.

New Jersey man charged with creating and posting deepfake nudes of two women. Bergen County prosecutors charged a 37-year-old Old Tappan man on October 22 with creating and distributing deceptive visual media after AI-generated nude images of two adult women were posted alongside their real names and photographs, Patch reports.

The pattern

1. The most striking synthetic political content this week came from candidates' own accounts rather than anonymous trolls or foreign networks. A county judge's campaign page posted fabricated supporters, a prime minister posted an AI-generated photo on polling day, and in Mississippi a Senate nominee's undisclosed AI ad was taken down only under threat of litigation. Election deepfake rules were largely written to stop someone from impersonating a candidate, and they have far less to say when the candidate is the one posting.
   ‍
2. Investment fraud shows the same tools working at scale against the people least equipped to question them. UK police counted more than 1,300 AI-enabled investment fraud reports in eight months, nearly 80 percent from people over 60, and much of that fraud runs through the same ad infrastructure where our own scan found AI-generated creative almost never disclosed. Awareness campaigns and firm registration checks help, but they ask the most targeted people to do the verification work that the distribution channel is not doing. The agent impersonation campaign shows the same weakness moving into machine traffic, where a claimed identity is trusted until someone checks how it actually behaves.
   ‍
3. Queensland's bill points in a useful direction, toward the moment of creation rather than the moment of harm. Distribution-only laws, platform takedowns, and fact-checks all act after synthetic content is already out in the world. Rules that attach to generation, whether criminal penalties for creating intimate deepfakes or provenance signals embedded by the tools that produce synthetic media, could set the tone for how other jurisdictions think about accountability before the damage is done.

Watching next week

• US midterms on November 3. Whether the late wave of campaign-posted AI content in Texas, Mississippi, and elsewhere produces any further complaints or enforcement before polls close.
• Israel's foreign interference probe. The head of the Central Elections Committee has asked the Shin Bet for findings within a week, as coalition negotiations begin after an inconclusive result.
• Queensland's deepfake creation bill. How quickly the bill moves through parliament, and whether other Australian states follow with creation offences of their own.
• Delhi High Court on Kriti Sanon. The court has said it will issue a comprehensive order on how personality rights extend to AI-generated likeness content, which could become the reference point for similar cases in India.

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.