The Deepfake Watchlist: Week of June 26 – July 2, 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.

1. A senior speaks out about the AI Carney scam that cost her $900K

CTV News reported this week, in the first public account of her ordeal, that 86-year-old Judy Skene of Sault Ste. Marie, Ontario, lost more than $900,000 to a crypto investment scam that began in the summer of 2025 with a Facebook ad featuring a deepfake video of Prime Minister Mark Carney.

• Category: Fraud / Impersonation
• Type: Attack
• Modality: Video, Audio
• Policy / Regulatory: Canada has no federal law specifically requiring platforms to detect, label, or pre-screen AI-generated deepfakes used in financial advertising.
• Trend: Deepfake videos of sitting heads of government and central bankers endorsing crypto platforms have become one of the most common vectors in fraud targeting elderly investors, with Carney joining former PM Justin Trudeau as a recurring face in these scams.
• Attack vector: A Facebook-hosted video ad using a cloned likeness and voice of a recognizable political figure to lend false institutional authority to a fraudulent investment platform.

What makes this case land harder than most fraud stories is the specificity of the false authority being borrowed. Skene didn't just see a stranger promising returns, she saw a Facebook ad appearing to show her own prime minister personally vouching for a $350 investment backed by the Bank of Canada, and that borrowed legitimacy is what got her past the skepticism a cold pitch would have triggered. Over the months that followed, she mortgaged her condo for $300,000 and cashed out her retirement fund before the account, and the money, disappeared.

The unresolved question is who bears responsibility for the ad running in the first place. Skene's advocate has argued platforms profit from hosting these scams as paid advertising, which puts accountability on Meta's ad review process, not only on the fraudsters who built the video. Watermarking at the point of generation would not have stopped this ad from being made, but it could have given Facebook's review systems a verifiable signal to catch before it ever reached Skene's feed.

2. An Illinois governor's race gets its first AI-native campaign

Capitol News Illinois reported that Republican gubernatorial candidate Darren Bailey and running mate Aaron Del Mar have made AI-generated images and videos a regular feature of their campaign, prompting Governor JB Pritzker to call for disclosure and watermarking requirements on political advertising.

• Category: Political / Electoral
• Type: Attack
• Modality: Image, Video
• Policy / Regulatory: A bill that would have required AI disclosure in Illinois campaign ads ahead of the 2026 election did not pass before the mid-June deadline.
• Trend: Campaigns are increasingly using openly labeled, cartoonish AI content rather than attempting convincing deepfakes, testing a different kind of persuasion built on volume rather than realism.
• Attack vector: Low-cost AI image and video generation used to produce a steady stream of satirical attack content aimed at an opponent, sidestepping the expense of traditional campaign media production.

What's interesting about the Bailey campaign's approach is that none of it tries to deceive anyone in the traditional deepfake sense, since the images are cartoonish and obviously synthetic, showing Pritzker in a Packers jersey or perched on a toilet-shaped throne. That distinction matters less than it might seem, because the volume and normalization of AI content in a campaign is its own kind of shift, regardless of whether any single image is realistic enough to fool a voter.

Pritzker's response is the more relevant part of the story. Rather than treating the content as beneath a reply, his campaign used it to argue for mandatory watermarking and disclosure requirements, a position he framed around AI "deteriorating the political environment." A bill requiring exactly that missed its deadline before this cycle began in earnest, which means whatever rules eventually apply to the 2026 election could arrive after most of the content has already been seen.

3. A Bondi Beach survivor's injuries were faked twice

The Times of Israel reported that Arsen Ostrovsky, wounded in the December Hanukkah terror attack at Bondi Beach, testified before Australia's royal commission on antisemitism that AI-manipulated images falsely depicted his injuries as staged.

• Category: Harassment / Public Safety
• Type: Attack
• Modality: Image
• Policy / Regulatory: Australia's federal royal commission on antisemitism and social cohesion, formed after the attack, is currently hearing testimony on social platforms' role in spreading this kind of content.
• Trend: Survivors of mass-casualty events are increasingly targeted with AI-manipulated images designed to recast real injuries as staged, sometimes called "crisis actor" content.
• Attack vector: A private selfie sent to a friend after the attack was distorted with AI to alter or remove visible injuries, then recirculated to accuse Ostrovsky of faking his wounds.

There's something particular about the cruelty of this category of attack, since it doesn't just spread a lie about a public figure, it retraumatizes someone who survived a mass shooting by accusing him of lying about his own injuries. Ostrovsky described the effect plainly to the commission, saying the content tries to "completely erase" his experience and his trauma, a different kind of harm than the reputational damage most deepfake stories focus on.

The platform response split along familiar lines. Meta removed the false content quickly once notified, while Ostrovsky said he received no response from X or YouTube, and the commission's counsel noted that other platforms, naming Gab specifically, have been openly hostile to cooperation requests. That gap between platforms willing to act and platforms that treat moderation requests as an imposition is likely to keep surfacing as more of these inquiries proceed.

4. A K-pop label's deepfake crackdown gets a status update

kbizoom reported that Source Music, the agency representing LE SSERAFIM, announced via Weverse that it has submitted a formal request for maximum punishment against creators of AI-generated sexual deepfakes targeting the group and will not accept settlements in ongoing cases.

• Category: CSAM / NCII
• Type: Response
• Modality: Image, Video
• Policy / Regulatory: The statement comes as South Korean authorities continue a nationwide crackdown on cyber sexual crimes running from last November through this October.
• Trend: K-pop agencies have moved from reactive takedown requests toward sustained legal campaigns, including internal monitoring systems built specifically to trace anonymized distribution.
• Attack vector: AI-generated sexually explicit content synthesizing group members' faces onto other bodies, distributed through closed platforms like Telegram previously treated as difficult to trace to a real distributor.

In Source Music's statement the agency says it has refused settlement offers even when perpetrators tried to pay their way out of prosecution. That's a meaningfully different posture from the quieter takedown requests that used to define how agencies handled this problem, reflecting a broader shift toward treating deepfake sexual content as a prosecutable crime rather than a moderation issue.

The claim that anonymous distribution on platforms like Telegram is traceable is worth taking seriously, since Source Music says it has identified and pursued individuals through exactly that kind of channel, with sentences already handed down. If that holds up as a repeatable pattern, it changes the calculation for anyone assuming that closed or encrypted platforms offer real protection from prosecution, an assumption that has driven a lot of this harm.

5. A model's face, minus his identity

TheGrio reported that Nigerian-Australian model Elii Emeghebo has filed a racial discrimination complaint against menswear brand Peter Jackson Australia after discovering the company used AI to lighten his skin tone and reshape his features in its advertising.

• Category: Brand / Likeness
• Type: Attack
• Modality: Image
• Policy / Regulatory: The complaint is filed with the Australian Human Rights Commission; the brand has acknowledged using AI to alter Emeghebo's image while disputing that race was a factor.
• Trend: A growing number of models are discovering their contracted likeness was AI-altered without consent after the fact, following a similar May case involving Dominican model Francheska Pujols and Rainbow USA.
• Attack vector: A brand used AI to modify a model's contracted image beyond the terms of the original agreement, altering protected characteristics without additional consent or compensation.

Emeghebo's description of the moment he found the altered image makes you pause as he described walking past his own advertisement and seeing "basically you, but without your identity there." That's a distinct kind of harm from the fraud and harassment cases in this issue, one where synthetic content isn't fabricating an event so much as quietly erasing the person who was actually hired and photographed.

The compensation question sitting alongside the discrimination complaint is also interesting sinceEmeghebo's legal team argues he should have been paid for the use of an AI-generated image his contract never anticipated. That's a dispute likely to recur across the modeling industry as brands treat AI alteration as a routine post-production step rather than a new use of someone's likeness.

Honorable mentions

Law Roach Reveals How Zendaya's Wedding Dress Compares to Viral AI Photos — E! News. AI-generated photos of Zendaya and Tom Holland's wedding at Lake Como surpassed 10 million likes when they first circulated in March, and stylist Law Roach put the rumor to rest again this week on Good Morning America. It's a reminder that a debunked deepfake doesn't necessarily stay debunked, since a single new interview can send months-old synthetic content back into circulation.

Former teacher with Kannapolis City Schools charged with child sex crimes, court records show — WBTV. A former Kannapolis, North Carolina middle school teacher was charged with third-degree sexual exploitation of a minor after police say he possessed an AI-generated image of a minor in sexual activity, adding to a growing list of prosecutions treating AI-generated CSAM with the same seriousness as photographic material.

The pattern

1. This week's stories split cleanly along a line that's becoming familiar: attacks that exploit a gap in a platform's or a jurisdiction's oversight (Carney's ad ran on Facebook with no disclosure requirement, Bailey's campaign content arrives ahead of any Illinois AI-ad law) and responses that show enforcement is possible once an institution decides to prioritize it (Source Music's traceability claims, the royal commission's platform testimony). The gap between the two says less about the technology's inevitability and more about which institutions have chosen to act.

2. Carney's story and Ostrovsky's story look different on the surface, one is financial fraud and the other is targeted harassment of a survivor, but they share a mechanic worth naming directly: both attacks borrow an existing, real moment, a public appearance, a private selfie, and use AI to make it say something the original never did. Provenance infrastructure at the point of capture or generation is the intervention that would blunt both, since it addresses the shared vulnerability rather than treating fraud and harassment as unrelated problems needing separate fixes.

3. The EU AI Act's Article 50 transparency obligations take effect in August, and this week's Illinois disclosure debate is a preview of the same fight playing out state by state in the US without federal coordination. Whether platforms build disclosure into their own systems or wait for legislatures to force the issue could determine how many more Carney-style scams run their course before anyone is required to label them.

Watching next week

• EU AI Act Article 50 enforcement. Transparency obligations for AI-generated content take effect in August, and the enforcement posture in the first weeks will be worth tracking.
• Illinois AI campaign disclosure legislation. Whether lawmakers revive the bill that missed this cycle's deadline now that Bailey's AI content has drawn national attention.
• Peter Jackson Australia complaint outcome. Whether the Australian Human Rights Commission treats AI alteration of protected characteristics as a novel category of discrimination claim.
• Royal commission testimony on antisemitism. Further hearings are expected to address platform cooperation, following Gab's public refusal to comply with commission requests.

‍

The Deepfake Watchlist publishes every Friday. 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 2025 Deepfake Threat Report.