Full Breakdown
The Global Push to Spot, Label and Regulate AI-Generated Deepfakes
7/12/2026, 11:11:55 AM
Core Event
Across academia, industry and government, a coordinated effort is underway to combat the surge of AI-generated synthetic media. Researchers have demonstrated that brief training can raise human accuracy in spotting AI-fabricated faces from roughly 40 % to 80 %. At the same time, policymakers in Australia are proposing mandatory disclosure and watermarking of AI-generated political content, while major platforms such as Meta and Spotify are rolling back or reshaping features that automatically reuse public images or AI-generated music.
Background & Context
The problem escalated after deepfake scams cost £12 bn in 2023, with Deloitte forecasting U.S. losses could climb to £40 bn in 2024. High-profile incidents include a Hong Kong-based employee transferring £25 m after a video call with a deepfake of their boss, and a 2019 AP investigation that linked a fabricated LinkedIn profile of “Katie Jones” to Russian intelligence.
Data & Statistics
- Participants in the University of Aberdeen study improved detection scores from ~40 % to ~80 % after a single hour of exposure.
- A 2025 study cited by Dr Caroline Azionya claimed only 1 % of people correctly identified deepfakes.
- Deloitte’s 2024 report projects U.S. deepfake-related fraud losses of £40 bn, up from £12 bn in 2023.
Official Statements & Responses
- Meta announced the removal of its “Muse Image” feature after backlash, stating: “Our intent was to provide a useful creative tool and to give people control over whether their public content could be referenced in this way. We’ve heard the feedback that this feature missed the mark, so it’s no longer available.”
- Deloitte warned that AI-deepfake scams could reach £40 bn in U.S. losses next year.
- Australian politician (unnamed in sources) is pushing legislation that would require disclosure and watermarking of AI-generated political material.
- Spotify introduced an “impersonation policy” in September 2024, mandating consent for any track that uses AI voice clones of an artist.
Criticism & Opposition
- Haley McNamara, executive director of the National Center on Sexual Exploitation, said the original Meta feature “obviously erode[s] our rights to our own likeness… but it is an obvious tool for #sextortion and other scammers!”
- Katrina German, founder and CEO of Ethical Digital, warned that automatic reuse permissions “make almost everything that you’re creating, unless you opt out, public domain.”
- Experts highlighted the “weaponisation” of synthetic media at the Social Media Summit for Government in Johannesburg, noting the difficulty of detection and the risk of misinformation.
Conflicting Reports & Gaps
- While the Aberdeen study reports a post-training accuracy of 80 %, Dr Azionya’s 2025 citation suggests a global identification rate of only 1 %, indicating a gap between controlled experimental settings and real-world performance.
- No concrete timeline has been provided for the Australian disclosure-watermark legislation, leaving its implementation date uncertain.
Verbatim Quotes
- “Training on visual artifacts, like looking for a sixth finger or odd earrings, has had limited success, partly because the AI is getting too good, and fraudsters may avoid using pictures with obvious flaws anyway,” — Prof Amy Dawel, Director, Australian National University Emotions and Faces Lab
- “That's helpful right?” — Dr Clare Sutherland, Psychologist, University of Aberdeen
- “People don't like their images being used in a way that they haven't given permission for,” — Katrina German, Founder & CEO, Ethical Digital
- “Muse Image includes robust technical protections designed to block the generation of policy-violating content,” — Meta spokesperson
What’s Next
- Meta’s blog indicates that future updates will focus on “robust technical protections,” though no release schedule is given.
- Spotify’s partnership with Universal Music Group will continue to expand AI-remix tools, subject to its September-2024 impersonation policy.
- The Australian proposal for AI-content disclosure is slated for parliamentary debate, but a specific vote date has not been announced.
These coordinated actions illustrate a growing recognition that AI-generated deepfakes pose tangible risks, prompting a blend of education, technical safeguards, and emerging regulatory frameworks worldwide.
