NewsResearch2026Global
The deepfake harms that arrived were not the ones detectors were built for
No date set
The dominant realised harms from 2022 to 2026 are peer-generated non-consensual intimate imagery, voice-clone scam calls aimed at families and finance staff, and emotional manipulation fraud. The large public-figure deepfake catastrophe that detection was built and funded for did not arrive.
Assess: Check whether your detection and your reporting flows match the harms people actually file, not the ones in the roadmap.
Who it reaches
It reaches user-to-user, marketplace, messaging and 6 other service types anywhere, at any size. Any service where people can be depicted or impersonated.
Lands on Policy & Standards, Product and Data & Analytics.
Learn more
Read the sourcePublished research
Not legal advice. Read the source.