Pangram, a three-year-old startup built by former Stanford researchers, just released its fourth AI text detector, called Pangram 4, and paired the launch with a nine million dollar funding round led by Menlo Ventures. That combination tells you something: detecting AI writing is turning into a real business, not a side project.
The company says its new model is over six times larger than the last one and cuts wrong accusations against human writers to roughly one in 24,000 documents. It also claims a sharp drop in missed AI text, and says it can spot AI writing run through "humanizer" tools, the software people use to disguise machine-written text as human, about 99 percent of the time across 13 commercial humanizer products.
Those numbers matter because AI detectors have a rough track record. Students have been wrongly accused of cheating because of detector errors, and research has repeatedly shown that tools like Turnitin carry false positive rates as high as four percent, with some detectors misfiring on non-native English writing far more often than that. A wrong accusation is not a minor bug: it can cost a student a grade, a job candidate an interview, or a freelancer a contract.
Pangram is positioning itself as the accurate alternative in a crowded field that includes GPTZero, Turnitin, Originality.ai, Copyleaks, and Winston AI. Its client list, according to the company, already includes Substack, which uses Pangram to show readers which newsletter writers use AI, along with Quora, universities, publishers, and recruiting firms. That is a useful list for any business leader wondering where this technology already touches their world: hiring pipelines, content platforms, and academic credentials all run through tools like this now.
The part of this launch that deserves more attention than the accuracy claims is the pricing change. Pangram is moving from a system where scans rounded up to blocks of 1,000 words to one where a scan uses up credits every 100 words. For short documents, that roughly doubles the effective cost. For long documents scanned through the API, the price can climb by up to ten times. Any company already paying for AI detection at scale, whether for reviewing job applications, checking vendor proposals, or moderating user content, should recheck their bill before the new pricing kicks in.
The bigger picture is that AI-written text is now everywhere, and businesses that depend on trusting written material, from insurance claims to academic transcripts to online reviews, will keep needing some way to check what is real. Pangram's claims are based on its own internal testing, not an outside audit, though an earlier version was independently reviewed by researchers from the University of Houston, UC Berkeley, and UC Irvine. Treat detector output as a strong signal to investigate further, not as final proof, no matter which vendor you use.