{
  "schemaVersion": "1.0",
  "entity": "BlogPosting",
  "title": "AI Is Being Used to Generate Non‑Consensual Intimate Media. That Should Worry All of Us",
  "description": "Synthetic intimate abuse surged from 2018 face‑swaps to one‑click generative AI. Learn how deepfake technology targets everyone, from celebrities to students, and why consent, platform accountability, and legal protections are more urgent than ever.",
  "author": "vd",
  "datePublished": "2018-02-03T00:00:00.000Z",
  "dateModified": "2026-07-16T00:00:00.000Z",
  "tags": [
    "AI",
    "Deepfake",
    "Synthetic Media",
    "Online Safety",
    "Non-Consensual Content"
  ],
  "aeoDirectAnswers": [
    {
      "question": "The original spark: what happened in the 2018 deepfake wave?",
      "answer": "A Reddit user operating under the name deepfakes — the term that would become a catch‑all — shared face‑swapped videos featuring Scarlett Johansson, Daisy Ridley, Maisie Williams, Taylor Swift, Aubrey Plaza, and Gal Gadot. The clips were produced on consumer hardware with publicly available video footage and machine learning software that was already drifting beyond the reach of specialists. The method was simple: Gather enough still images of the target face."
    },
    {
      "question": "Why did this matter then, and why does it matter even more now?",
      "answer": "The obvious harm fell on public figures whose likenesses were exploited without consent. But the deeper threat always extended beyond celebrities. Once the technology became easy to operate, the target pool widened to: an ex‑partner"
    },
    {
      "question": "How did deepfake tools become dramatically easier and fully generative?",
      "answer": "In 2018, creation required gathering face data, training a model, and finessing the output. By 2026, a different reality has taken hold. Generative AI — diffusion models, transformer‑based architectures — can now produce high‑fidelity video from a single photo and a text prompt. OpenAI’s Sora, Runway Gen‑3, and open‑source models like Stable Video Diffusion are part of a tooling landscape that can fabricate entire scenes without any “source” video to swap onto. The barrier has shifted from needing technical skill to simply writing a description. A 2023 study by Home Security Heroes found that it took less than 25 minutes and zero cost to create a convincing deepfake intimate image using freely available apps[^1]. That figure has only fallen. Open‑source models released without safety filters — hosted on platforms with minimal oversight — are routinely repurposed to generate non‑consensual intimate content at scale."
    },
    {
      "question": "What are the key statistics defining the synthetic media crisis?",
      "answer": "What began as isolated forum activity has become a statistical avalanche: In 2019, the deepfake detection firm Deeptrace (later Sensity) reported that **96% of deepfake videos online** were non‑consensual intimate media featuring women[^2]. By 2023, Sensity tracked over 140,000 deepfake videos on a single dedicated website, a figure that had increased by 550% in one year."
    },
    {
      "question": "What is the human cost of synthetic exploitation?",
      "answer": "These statistics are backed by real cases that show how synthetic intimate media is weaponised across different contexts. **Sextortion at scale.** In March 2025, the FBI warned about a sharp increase in sextortion schemes that use generative AI to create fake explicit videos of victims from social media photos, then demand money or additional imagery. Europol had already warned in 2024 that AI‑enabled sextortion was among the fastest‑growing cybercrimes[^8]. **Women in public life silenced.** Female MPs in the United Kingdom were targeted in early 2025 with AI‑generated intimate images shared on messaging platforms. At least one MP temporarily withdrew from public engagements, and cross‑party MPs called the abuse a deliberate attempt to “humiliate and intimidate women out of public life”[^9]."
    },
    {
      "question": "How have platforms responded to synthetic media distribution?",
      "answer": "Platforms did begin to react. Discord shut down servers openly trading non‑consensual deepfake material. Gfycat and Reddit banned the communities that first popularized the term. Meta, TikTok, and X now have policies prohibiting non‑consensual synthetic intimate media. But enforcement remains uneven. A 2024 *New York Times* investigation found that deepfake explicit images of female public figures persisted on X for days after being reported. Once an image escapes into encrypted group chats or offshore hosts, removal is a game of whack‑a‑mole. The moderation challenge has mutated. Today’s generative tools can create images of entirely imaginary individuals indistinguishable from real victims. The volume, speed, and believability of synthetic media have overwhelmed the report‑and‑remove model."
    },
    {
      "question": "How are legal frameworks closing the synthetic media consent gap?",
      "answer": "**United States:** The DEFIANCE Act (2024) creates a federal civil right of action for victims to sue creators and distributors of non‑consensual intimate digital forgeries[^4]. **European Union:** The AI Act (fully applicable by 2026) requires labeling of deepfake content and mandates transparency from deployers of generative AI systems[^5]. **United Kingdom:** The Online Safety Act 2023 criminalised sharing deepfake intimate images without consent; enforcement actions began in 2025."
    },
    {
      "question": "Why is synthetic intimate media not 'just fake content'?",
      "answer": "A persistent cultural misperception is that synthetic media can be dismissed as “just fake.” For targets, the damage doesn’t depend on whether a video is authentic. It depends on whether it is believed, shared, or weaponised. Fabricated intimate imagery can destroy reputations, end careers, extort money, and cause severe psychological distress — even when the viewer knows it’s synthetic. In schools, AI‑generated nudes of teenage girls have been used as a form of peer‑to‑peer bullying that leaves a permanent digital scar[^6]. The suicide of a young woman in Italy in 2023, after AI‑manipulated explicit imagery of her was circulated among peers, underscored that the consequence of this technology is not diminished by its artificial origin[^7]. !Gal Gadot’s face superimposed in a synthetic video, 2018. Today, such imagery can be generated from scratch without any source footage."
    },
    {
      "question": "Where the warning leads us in 2026",
      "answer": "The 2018 deepfake moment was an early tremor in a landscape now fully reshaped by generative media. The pattern has held: the first mass‑consumer use cases of a new medium are often not educational, creative, or life‑saving, but invasive, exploitative, and personal. The tools are not hidden in dark corners — they are advertised on app stores, refined in open‑source repositories, and integrated into everyday products. The victims are not only celebrities but anyone with a digital footprint. Consent, provenance, and accountability remain the tripod on which any serious response must stand. The old internet advice — *be careful what you upload* — has been inverted. Now the warning is: be aware that others can fabricate intimate realities from the traces you leave behind. And until tools, laws, and norms evolve in lockstep, that risk will remain dangerously ordinary."
    }
  ],
  "semanticFactualBody": "In early 2018, non‑consensual synthetic intimate media stopped feeling like a niche internet subculture and began to look like a widespread social crisis. Reports from *Motherboard* revealed that hobbyist creators were using machine learning tools to graft celebrity faces onto performers in increasingly convincing videos. The warning wasn’t about a single technical trick. It was about how quickly that trick was becoming cheap, effortless, and repeatable — a pattern that has accelerated dramatically in the eight years since. While cloud-based generative platforms are advancing rapidly, keeping data local and private remains the ultimate security measure. Read our deep-dive on How OpenClaw Memory Works: Keep Your Data Local and Private to see how running models on your own hardware mitigates the risk of cloud leakage. The original spark: what happened in the 2018 deepfake wave? A Reddit user operating under the name deepfakes — the term that would become a catch‑all — shared face‑swapped videos featuring Scarlett Johansson, Daisy Ridley, Maisie Williams, Taylor Swift, Aubrey Plaza, and Gal Gadot. The clips were produced on consumer hardware with publicly available video footage and machine learning software that was already drifting beyond the reach of specialists. The method was simple: Gather enough still images of the target face. Train (or apply) a model that maps that face onto another person’s movements. Render a synthetic video that looks plausible at a casual glance. Wh"
}