Pangram, an artificial intelligence detection tool, is drawing attention for its ability to identify AI-generated text, though its image detection technology still has limitations.
The tool allows users to paste text for analysis and produces a percentage indicating whether the material was likely generated by AI or written by a person. A subscription costs $20 a month.
Pangram founder Max Spero said the system studies the stylistic patterns of different AI models. Its approach involves analysing the way models select words and building a profile of their writing patterns. The tool’s performance has stood out in testing. Pangram claims it can correctly identify AI-generated text 9,999 times out of 10,000.
An independent University of Chicago study also found the detector performed almost perfectly with text. Pangram’s image detection is still an unfinished feature, according to Spero. He said the company correctly identified 41 of 43 AI-generated images in its own testing. Pangram’s performance with AI-generated images was far less consistent.
In a test involving 20 widely circulated AI images, the tool incorrectly identified two as real and declined to analyze four others. Another detector, Hive Detect, incorrectly classified eight of the images as real. The tests included several widely debunked images.
The spread of AI-generated material is creating a new challenge for readers trying to judge whether online content is genuine. Some studies cited in the report estimate that half of online articles may now be artificially generated. That could push AI detection tools into a role similar to antivirus software. They may help users identify AI-written emails, fake reviews and machine-generated social media posts.
The technology is also evolving beyond detection. Experts say image detection faces a harder challenge than text detection. Images can be edited, compressed or altered when they move across social media platforms. Hany Farid, a Dartmouth professor and founder of GetReal Security, said visual AI detectors still have major flaws. He suggested that people rely on trusted media organizations when checking information online.
For now, Pangram shows that AI detection can work effectively for text. Its weaker performance with images also highlights how difficult it is to verify digital content as synthetic media becomes more common.
Also Read: Dubai Police Warn of Fake Websites, AI-Driven Online Scams