Pangram, a New York startup, has launched two new models for detecting AI-generated images and video. Pangram Image 1.0 checks still images, while the faster Image Flash 1.0 analyses video frame by frame. The company claims its models reach near-perfect accuracy in its own tests and 98.7% on the DF26 academic benchmark for video. Both models are now available in Pangram’s web app and API, with Image 1.0 also in its Chrome extension.
The image model can show heatmaps of AI content in photos and reportedly catches 99.2% of AI images, with a 0.27% false positive rate. Pangram’s own test used 100 AI-generated copies of real photos, with Image 1.0 scoring 99.96% accuracy. The video model, Flash 1.0, outperformed other detectors on DF26, a benchmark where both humans and leading tools have struggled.
The tools now work on smaller images and are designed to see through compression and noise. Pangram’s Chrome extension flags AI content on social media and e-commerce sites. The company raised $9 million in July to develop these models. However, AI detectors are not foolproof: Pangram’s text detector has made errors, and even Meta’s tools have missed fakes in tests.
Banks, insurers and healthcare firms are exposed to fraud and misinformation risks from AI-generated content. As platforms increase scrutiny, firms may need to review detection options and update controls.
