AI detector Pangram 3.0 can detect text created, revised or edited using AI tools

Pangram, an AI transparency provider, has launched Pangram 3.0, offering users the ability to detect content that has been edited or modified by AI, as well as content created entirely using AI.

The upgraded tool provides enhanced AI detection, including if any section was created, assembled, or modified using AI tools, allowing educators to accurately verify a paper.

Pangram 3.0 was trained on hundreds of editing prompts for AI tools, including: “make this more descriptive” or “make this more casual”. The tool is capable of labeling sections of text as lightly or moderately AI-assisted, making it easier to make decisions around AI use.

“Pangram is the first AI detector to give insight into the subtleties that go into writing. In many sectors, especially in education, AI text detection was limited to a binary response: is there AI content or not?” explains Max Spero, Co-founder and CEO at Pangram. “Pangram 3.0 is the first detector that can show where a text was AI-generated, where it was fully human-written, and that in-between space where a text was co-authored with AI.

“Previous versions of Pangram were highly accurate at differentiating between human and AI, but could not tell you the extent to which AI assisted with a particular section. Now we can go deeper and paint a real picture of what’s going on. It’s a significant step forward for our system and for the field of AI transparency.”

Earlier this year, Pangram raised an additional $2.7 million in its latest seed round, taking the total raised to nearly $4 million. Pangram plans to use the funding to continue developing partnerships with higher education institutions, which have been struggling to address misuse of AI among their students.