Pangram raises $9M to detect AI text and images
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Pangram raised $9M led by Menlo Ventures with participation from Haystack, ScOp, Script Capital, and Cadenza, alongside the launch of its next-generation text detection model and an image detection model in research preview
- Pangram 4 claims over 99% accuracy at flagging AI-assisted writing and mixed human-AI text, and is better at detecting AI humanizer programs, with a false-positive rate of roughly 1 in 10,000 human documents
- Pangram Image works on pixel-level statistical distributions to detect AI-generated images across multiple models, positioning itself as an alternative to OpenAI's and Google DeepMind's watermark-based checks, which mostly detect their own output
- Substack integrated Pangram's API to show readers when authors use AI for newsletters; Quora, schools, universities, publishers, agents, and recruiters are also API customers
- Max Spero and Bradley Emi, Stanford AI and machine learning grads, founded Pangram about two years ago after ChatGPT's launch, training its model on tens of millions of human documents paired with "synthetic mirrors" written by frontier LLMs
- Competitors Winston AI, Originality.ai, Copyleaks, and GPTZero are also building AI detection tools as institutional pushback grows, with arXiv introducing a one-year submission ban this year for unreviewed LLM output
- TechCrunch hands-on testing found Pangram easily flagged AI-generated articles from ChatGPT and Claude and resisted humanizer prompts, though it occasionally mislabeled human-rewritten text as AI-assisted
Why it matters: Pangram's $20/month subscription and Chrome extension that auto-labels posts on X, LinkedIn, Substack, Reddit, and Medium positions the startup as a real-time authenticity layer for the social web. Substack's integration is the clearest signal that platforms—not just individuals—are willing to pay for AI provenance, and the cross-model image detection claim addresses a gap that single-vendor watermarks from OpenAI and Google DeepMind haven't closed.
