Anthropic's 335-page IPO prospectus tested by Pangram: 92% judged as human-written
Business Insider fed the full text of the 335-page IPO prospectus submitted by Anthropic into the AI writing detection tool Pangram, and the results showed that 92% of the roughly 259,000-word document was judged to be human-written, while some sentences and paragraphs were flagged with "high" confidence as containing traces of AI use. This result is only the judgment of the detection tool, and it cannot be confirmed whether Anthropic used its own AI assistant Claude. Anthropic did not respond to a request for comment, and its AI watermarking system has not yet been made available to ordinary users.
Body · AI translation
On October 27, according to Business Insider, the artificial intelligence company Anthropic submitted a 335-page IPO prospectus. To find out whether the document was written by AI, Business Insider fed the full text into the AI writing detection tool Pangram for analysis.
Pangram's detection results showed that 92% of the roughly 259,000-word document was judged to be human-written, while some sentences and paragraphs were flagged with "high" confidence as containing traces of AI use.
However, this result is only the judgment of the detection tool; it cannot prove which specific content was written by humans or AI, nor can it confirm whether Anthropic used its own AI assistant Claude.
The Business Insider reporter initially tried to test the entire document herself but ran into difficulties because the file was too large. Subsequently, Pangram co-founder and CEO Max Spero helped complete the full-text detection.
Previously, a Business Insider reporter had entered her own LinkedIn posts and published articles into Pangram. The tool flagged one sentence as AI-written. Reviewing her writing process, she found that she had originally written that sentence and later accepted a rewriting suggestion from ChatGPT.
Anthropic did not respond to Business Insider's request for comment on the detection results. The outlet said Anthropic's own AI watermarking system has not yet been made available to ordinary users, so it was not used for this detection.
Business Insider author Katie Notopoulos believes the result raises an interesting question. As an AI company, why did Anthropic not make more use of its own products to write the prospectus? But she also noted that a prospectus must undergo review by the U.S. Securities and Exchange Commission, every sentence in the document may come under scrutiny, and having humans involved in writing it also has its rationale.