Deepfakes and impersonation

The U.S. Department of Justice and Homeland Security Investigations seized the CFAKE.com and SOCFAKE.com websites, which allegedly hosted nonconsensual AI-generated nude images and videos of women, in what appears to be the first publicly announced domain seizure under the TAKE IT DOWN Act. The seizure was part of a coordinated operation with Italy and France, and a suspect was arrested in Nice, France, on June 10. The TAKE IT DOWN Act makes publishing such intimate imagery and digital forgeries without consent a federal crime and requires platforms to remove reported images within 48 hours of a valid victim request.

A UK-based Chinese activist, Apple Peiqing Ni, founder of the China Dissent Network, was targeted on X by deepfake posts portraying her as a sexually promiscuous drug addict. UK police advised her to complain to the platform, but X told her the abuse did not breach its rules.

The Bank of England has warned the public against AI-generated scams after deepfake videos showing Nigel Farage fighting its governor, Andrew Bailey, spread on X. Bailey urged people to report the videos so they could be taken down, noting that fake adverts impersonating the Bank and other central banks are on the rise. The Bank has raised concerns about the posts with Reform UK and social media platforms.

xAI has asked a federal court in northern California to overturn an earlier ruling that let four plaintiffs in a class-action lawsuit use pseudonyms. The plaintiffs allege that Grok was used to create sexualized deepfake images of them, including one image of a child, and fear harassment if their real names are disclosed. xAI argues that the public interest in the identities of those suing outweighs their privacy concerns.

Researchers Shuai Wang, Gaobo Yang and Hanling Zhang published a deepfake detection study in the Journal of Information Security and Applications, Volume 101, in September 2026. The source text provided contains only the publication metadata and no description of the method or findings.

Researchers propose the Degradation Consistency Learning Framework to improve deepfake face detection on forgery methods unseen during training. The method exploits degradation inconsistencies between background and manipulated face regions, using a data generation network and a detection network mining both spatial and frequency domain clues, coupled through adversarial training. The authors report effectiveness on benchmark datasets under in-dataset and cross-dataset protocols.

Google is expanding AI content detection to Chrome and Search so users can check whether images were made or altered with AI. The updates cover SynthID, the invisible watermarking technology developed by Google DeepMind, and content carrying C2PA content credentials. Verification for SynthID-marked images is rolling out to Search features, including Google Lens, starting with the announcement at Google I/O.

Ofcom, the UK communications regulator, will update its codes of practice to require online service providers to detect and remove intimate image abuse, including AI-generated deepfakes. The change follows a rise in revenge porn and deepfakes targeting women and girls, and a wave in January when Grok AI was widely used to create sexualised videos of women in bikinis.

FaceReclaim is a diffusion-based framework that tries to restore the original face hidden inside a face-swapped deepfake image, a task the authors call Traceability of Face-swapping Deepfake (TFD). It reformulates TFD as image-to-image editing and uses Multi-Scale Face Attribute Decoupling and Multi-Modal Face Identity Prompting as dual conditioning. The authors report that synthetic faces retain subtle traces of the original face, and they evaluate restoration both visually and through face verification.

South Korea's local elections next month will serve as a test of how effective regulations can be at curbing the spread of deepfakes. The source frames the vote as a real-world trial of whether such laws work.

YouTube is expanding its AI likeness detection program to all users over 18. The feature uses a selfie-style face scan to search the platform for lookalikes and alerts the user when it finds a match. The user can then request removal of the content.

The authors introduce a taxonomy and benchmarking framework that classifies deepfake attacks by intent and generative mechanism, then evaluate them against deep face recognition systems. Diverse deepfake attacks commonly exceed 70% success and, in some regimes, surpass 90%. The study finds that attack success depends on the degree of identity controllability rather than visual quality.

PVLM is a parsing-aware vision-language model with dynamic contrastive learning for zero-shot deepfake attribution, meaning it traces forged faces to generators not seen in training, including diffusion models. The authors build a fine-grained ZS-DFA benchmark and use face parsing to exploit differences in how GAN and diffusion generators preserve source facial attributes. They report that the model exceeds the state of the art on the ZS-DFA benchmark across various protocol evaluations.

A newsletter collects several tech stories. Adult content creators say AI systems train on their work, clone their likenesses, and generate explicit content they never agreed to make. Separately, generative AI chatbots such as Gemini have surfaced people's real phone numbers, and experts believe the leaks stem from personally identifiable information in AI training data.

Jennifer, a psychotherapist who appears under a pseudonym, found that a facial recognition search of her new headshot returned an old adult video of hers with another person's face swapped onto her body. The article reports that porn performers' bodies are used as training data for AI-generated nudes and that their likenesses are re-created without consent, threatening their livelihoods.

Researchers propose a proactive Deepfake detection framework that embeds a watermark into the Quaternion Radial Harmonic Fourier Moments (QRHFMs) domain of face images. A Local-Enhanced State Space (LSS) block extracts features, and a dual-branch architecture pairs a watermark extractor with a forgery discriminator via knowledge distillation. On benchmark datasets the method reports 92.58% average accuracy and 95.96% AUC, with robustness to image quality degradation and to Deepfake techniques unseen during training.

A finance director at a Singapore multinational authorized a $499,000 transfer during a Zoom call where every participant was AI-generated, after the attacker had mapped the org chart and the wire approval workflow. The article says the same method stole $25.6 million from Arup in early 2024, and that voice deepfake incidents rose 680% year-over-year in 2025, with deepfake fraud losses exceeding $200 million in the first four months of 2025.