Anthropic's Documents Screened for AI Traces
Pangram detector recognized 92% of Anthropic's IPO application text as human-written
Business Insider journalist Katie Notopoulos decided to investigate whether Anthropic used artificial intelligence to prepare documents for its stock market debut. To do this, she submitted the company's 335-page registration statement to the Pangram service, which scans for signs of machine-generated text.
Processing such a large file independently was not feasible, so Pangram co-founder and CEO Max Spero assisted with the analysis. According to the results, the service classified 92% of the document as human-written text. In certain sentences and fragments, it detected potential signs of AI use.
Spero believes this outcome more likely indicates light editing or assistance in formulating specific sentences. He found no evidence that significant sections were entirely generated by a neural network.
However, the detector's assessment does not prove how the document was created. It also does not confirm that the remaining 8% was written by Claude: the service may make errors, and processing human text with AI can affect the results.
At the time of publication, Anthropic had not responded to the journalist's inquiry. Thus, the review remains an experiment with a detector rather than confirmation of the company's use of a specific tool.
