30-Minute Training Boosts AI Deepfake Detection Accuracy to 82%, ICWSM Study Shows
mattgroh · x · 2026-08-03
An upcoming ICWSM 2027 paper examined the impact of brief training on detecting AI-generated images. The study found that US government intelligence analysts improved their accuracy in distinguishing real photos from AI fakes from 73% (on par with the general public) to 82% after just 30 minutes of training.
The training highlighted patterns and tell-tale signs using 7 real and 50 AI-generated images. Furthermore, their ability to accurately classify both images in a pair, such as real and fake astronauts, increased by 11%.
More from Safety
- OpenAI Disrupts Cambodia-Based Criminal Scam Operation Using ChatGPT — OpenAI News · 2026-08-04
- GovAI Report: Could AI Enable Catastrophic Cyberattacks on the US Power Grid? — Manderljung · 2026-08-04
- MIT Tech Review: AI Agents 'Lying' Is Actually Reward Hacking — orbitalNest · 2026-08-04
- XCSSET Malware Variant Targets macOS Developers via Supply Chain — cyb3rops · 2026-08-04
- AI-Enabled Worms Could Cause Tens of Billions in Damage; $100B Threshold Too High — Manderljung · 2026-08-04
- AI Coding Agents Accused of Unauthorized Edits to Sensitive Code — bluetech333 · 2026-08-04