The Deepfake Watchlist: Week of October 30–November 5, 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: Russia's Storm-1516 rewrote real local newscasts to accuse Senate candidates of abuse

PolitiFact's Russian influence operation linked to attacks targeting Democrats in Maine, Ohio, Texas, published October 30, reports that the Russian influence network Storm-1516 used AI to alter genuine local TV news segments so they appeared to report fabricated abuse allegations against Ohio Senate candidate Sherrod Brown and Maine Senate candidate Troy Jackson in the final days before the midterms.

• Category: Political / Electoral
• Type: Attack
• Modality: Video, Audio
• Policy / Regulatory: PolitiFact, NewsGuard, and the Associated Press traced the videos to a known foreign influence network, but the reporting describes no takedown or enforcement action against the accounts that spread them before Election Day.
• Trend: Foreign influence operations moving from scripted actors to AI lip-sync edits of authentic news footage, borrowing the credibility of trusted local outlets.
• Attack vector: Real broadcast segments from Cleveland 19 News and WABI-TV re-voiced and lip-synced to deliver fabricated allegations, then amplified by a rotating cast of paid and unwitting influencers.
• What we saw in the content: Reporting on the videos documented the kinds of signals detection models are trained to flag, for example:
  ‍
  • The lip movements of Cleveland 19 reporter Nicole Vransky were altered to match words she never said, the signature of an AI lip-sync edit
  • The source footage underneath the edit was left intact, with Vransky wearing the same red blouse she wore in the genuine October 21 report on Ohio's governor's race, which let fact-checkers trace the fake back to its original
  • A real July WABI-TV report about Jackson throwing a water bottle as Maine Senate President was re-edited so it appeared to report that he threatened staffers with an axe
  • Former Brown staffers named in the video told PolitiFact and AP that the allegations attributed to them were false

Two weeks ago we featured the Storm-1516 video that used the identity of a real former student, Zyon Lewis, to accuse James Talarico of abuse. This round borrows something more valuable than a stranger's face, which is the authority of a local newsroom. Local TV news still carries a level of trust that national outlets have struggled to keep, and an AI lip-sync edit lets an operation rent that trust by reusing a reporter's real footage and changing only her mouth and her words. PolitiFact notes that in 2024 this network hired actors to read its scripts, and that it now uses AI instead, which brings the cost of each new fake close to nothing.

The reach numbers explain why that is enough. Clemson's Darren Linvill told PolitiFact that fewer than 7,000 people mentioned Troy Jackson on X on October 28, yet a quarter of them were discussing the Russian narrative. In a midterm race, a few thousand well-placed impressions can take over a large share of a small local conversation. Brown, Jackson, and Talarico all won their races on November 3, but the outcome does not undo the harm to the real people whose names and words were used, including former staffers who had to publicly deny accusations they never made.

Cleveland 19's original report exists and anyone can watch it. The problem is that nothing travels with the altered clip to point back to it. Provenance signatures applied by broadcasters at publication would let a viewer, or a platform, check whether a clip still matches what the station actually aired, and that check works the same whether the edit is crude or flawless. Until newsrooms sign their output by default, their archives are raw material for the next operation that wants to sound like the evening news.

1. A network of 400+ ordinary-looking AI apps was hiding sexual deepfake tools

WIRED's Hundreds of Mundane AI Apps Are Hiding Sexually Abusive Deepfake Features reports that more than 400 linked apps and websites, presented as pet-portrait generators, workout coaches, and language tutors, concealed tools for creating sexual deepfakes from uploaded photos, with Apple removing 19 apps after WIRED shared its findings.

• Category: CSAM / NCII
• Type: Attack
• Modality: Image, Video
• Policy / Regulatory: Apple says the apps met its rules at review and developers changed what they loaded afterward; Meta says it took action against 10,000+ ad accounts and blocked 209 apps from advertising.
• Trend: Nudify tools disguising themselves inside ordinary consumer apps to survive app store review and advertising policies.
• Attack vector: Apps that send a user's language, referral source, and identifiers to an external server, which decides whether to show the advertised interface or load the abusive one, often for users arriving from Facebook and Instagram ads.

Researchers at AI Forensics found 27 of the apps on Google Play in September with more than 350,000 combined downloads, and the abusive features appeared only when a remote server decided the visitor was the right audience. AI Forensics researcher Paul Bouchaud described it to WIRED as an organized, commercially run operation, and the security firm Infoblox linked it to nearly 500 related domains.

The ad system keeps showing up as the delivery mechanism. In September we covered the Tech Transparency Project's discovery of AI-generated abuse ads running on Meta, and here the ads were the trigger that unlocked the abusive interface. Meta's enforcement numbers are large, but WIRED reports that similar apps and ads kept appearing while the story was being reported. For the people whose photos end up in these tools, a removal after the fact matters far less than whether the tool was reachable in the first place.

2. Binance says a fake job interview opened the door to its $6.3 billion heist

Help Net Security's Fake job interview opened the door to $6.34 billion Binance heist reports that Binance's November 3 incident report traces the October 29 theft to a fake recruitment process, in which an engineer was interviewed by an impostor HR representative and a supposed CEO of a fictitious company, then asked to run a malware-laced coding assessment on his work computer.

• Category: Fraud / Impersonation
• Type: Attack
• Modality: Video
• Policy / Regulatory: The FBI has attributed the theft to North Korea's TraderTraitor group, and Binance says its findings will be subject to further independent review.
• Trend: State-backed actors using fabricated recruiter and executive personas, including live video screenings, as the opening move in large-scale intrusions.
• Attack vector: Fake project director on a legitimate job platform → staged HR and CEO interviews with a live video screening → malicious "assessment" repository on a company laptop → internal AI tool used to compromise two more engineers → wallet-signing service abused.

Binance has not said whether the people on that video screening were deepfakes, and we are not assuming they were. What the report does show is that the largest crypto theft on record began with a person on a screen who was not who they claimed to be. Binance also says it detected and stopped 57 suspected North Korea-linked attempts this year, including fake job interviews, poisoned developer tools, and fake video-call apps.

Identity checks in recruiting are mostly informal, a profile, a video call, a name that matches. That is the layer North Korean operators have industrialized, both to get hired through IT worker schemes and, in this case, to get an engineer to run their code. Binance admits the malware itself slipped past its endpoint protection, which is a good argument for treating the question of whether the face and voice on a recruiting call belong to a real, present person as part of the security stack.

3. Real footage of a fatal plane crash was dismissed online as "100%" AI

PolitiFact's Palm Beach plane crash footage isn't AI-generated, despite social media claims reports that X posts falsely claimed dashcam footage of the November 1 Learjet crash in Palm Beach County, which killed everyone on board, was entirely AI-generated, using screen recordings of an AI detection tool as supposed proof.

• Category: Harassment / Public Safety
• Type: Attack
• Modality: Video
• Policy / Regulatory: PolitiFact rated the claim Pants on Fire after the AP, CNN, CBS, and BBC carried the footage and PolitiFact traced it to the dashcam owner who first posted it; no platform action against the posts is described.
• Trend: The liar's dividend arriving at breaking news, where authentic footage of a real tragedy is dismissed as synthetic within hours.
• Attack vector: Viral X posts presenting a detection tool's output as evidence that real footage was fake, with at least one account amplified by bot accounts that promote Storm-1516 narratives.

Most of what we track is fake content passed off as real, and this is the reverse, which is just as damaging. Linvill told PolitiFact that people have become so used to being lied to that they no longer recognize the truth when they see it. When dashcam footage of a fatal crash is labeled synthetic within a day, families and investigators end up having to defend the fact that the event happened at all.

The posts leaned on screen recordings of a detection tool flagging the clip, and when PolitiFact ran the same footage through the same tool, it was not flagged. A detection score with no method, no context, and no record of who ran it can be turned into a prop as easily as a fake video. That is worrisome for anyone who builds or relies on detection, because the public's trust in these tools depends on results that come with an explanation people can check.

4. Delhi High Court grants Kriti Sanon interim protection against AI deepfakes and voice clones

LiveLaw's Delhi High Court Protects Actress Kriti Sanon's Personality Rights Against AI Deepfakes, Impersonation reports that Justice Purushaindra Kumar Kaurav granted the actor interim protection on November 4, ordering takedowns of identified URLs and restraining defendants from generating similar content, including sexually explicit deepfakes and AI chatbots or voice models presented as Sanon.

• Category: Brand / Likeness
• Type: Response
• Modality: Video, Image, Audio
• Policy / Regulatory: Intermediaries must disable the listed URLs within the deadlines set by India's IT Rules, 2021, authorities were directed to issue blocking orders for sites hosting pornographic deepfakes, and the next hearing is December 10.
• Trend: Indian courts extending personality rights doctrine to AI voice models, chatbots, and generated likenesses in the absence of a dedicated statute.
• Attack vector: Not applicable, this is a judicial response.

Last week we flagged that the court planned to issue a comprehensive order in this case. What arrived on November 4 is interim, but it names AI chatbots and voice models presented as the actor alongside deepfake imagery, impersonating accounts, and merchandise. Treating a likeness as something that can be infringed across every modality matches how this abuse actually shows up for public figures.

The order excludes news, reviews, criticism, satire, parody, caricature, and fan expression, and the court described its findings as prima facie, with the harder legal questions left for after the contesting defendants are heard. December 10 is where this could become precedent other Indian courts lean on.

Honorable mentions

Italian court orders Meta to filter deepfake investment ads. The Court of La Spezia issued an urgent order on October 29 requiring Meta Platforms Ireland to remove fraudulent ads using the face and voice of financial adviser Salvatore Gaziano, deactivate the linked WhatsApp groups, and put filters in place to stop the ads from reappearing, Il Sole 24 Ore reports. A hearing to confirm the order is set for December 10.

Former Pennsylvania state trooper sentenced for creating deepfake pornography. A Montgomery County judge sentenced former corporal Stephen Kamnik to five to 10 years in state prison on November 2 for creating thousands of deepfake pornographic images of about two dozen women, including co-workers, and for possessing child sexual abuse material, The Philadelphia Inquirer reports. He was designated a sexually violent predator and barred from working in law enforcement.

Arizona prosecutors appeal the Pelkey resentencing. Prosecutors are asking the Arizona Supreme Court to review the appeals court ruling that vacated a manslaughter sentence because an AI-generated video of the victim influenced the judge, KJZZ reports.

The pattern

1. This week's attacks all borrowed trust from an institution that was never designed to verify what passes through it. Storm-1516 borrowed a local newsroom's credibility, the nudify network borrowed the app stores' approval, and North Korean operators borrowed the legitimacy of a real job platform and a live video call. In each case the institution did its own job reasonably well, and the attack lived in the gap between what the institution checked once and what it delivered later.
   ‍
2. The Palm Beach crash shows what happens when that erosion of trust runs in reverse. Once enough synthetic content circulates, real footage becomes deniable, and a detection screenshot with no context becomes a prop for denial. Detection still matters, but it has to come with an explanation people can check, and authentic footage needs a way to prove where it came from that does not depend on anyone's word.
   ‍
3. Courts are moving faster than legislatures to fill those gaps. The Delhi High Court's order reaches AI chatbots and voice clones, an Italian judge ordered Meta to filter deepfake ads as well as remove them, and the EU's ban on nudifier apps is due to take effect December 2. Orders that require platforms to prevent reappearance, on top of answering complaints, could set the tone for how accountability works when the same synthetic content can be regenerated in seconds.

Watching next week

• Post-election narratives from Storm-1516. Linvill told PolitiFact he expected the network to push narratives undermining trust in the election on and after Election Day, now that the results are in.
• Follow-through on the hidden-app network. Whether Apple, Google, and Meta remove the remaining apps, domains, and ads WIRED says kept appearing during its reporting.
• Binance's independent review. Whether the outside investigation clarifies who was on the recruiting video call and how the fake personas were built.

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.