When Substack quietly flipped the switch on a "Scan for AI" button last month, the feature landed with barely a whisper. One click, and subscribers could gauge whether their favorite newsletter was human-crafted or algorithmic. Under the hood: Pangram Labs, a two-year-old Brooklyn outfit that believes it has cracked what may be the internet's thorniest content-authenticity puzzle.
Now Pangram has something else to announce. On July 29, the company disclosed a $9 million seed extension led by Menlo Ventures' Anthology Fund—capital raised, it says, to double down on detection accuracy and push into image analysis. The timing feels deliberate. Universities are backing away from AI detection tools. Students contest false accusations. And the industry Pangram is trying to corner increasingly resembles a minefield.
The round pulled in Haystack, ScOp Venture Capital, Script Capital, and Cadenza. Combined with a prior $3.98 million raised in June of last year, Pangram has now disclosed roughly $13 million in total funding. Valuation? The company declined to share.
The Substack Deal, and What Came Before
Substack's rollout—July 21, according to the platform's support docs and a TechCrunch write-up—marks one of the first times a major consumer platform has embedded AI detection at scale. Writers publishing after that date can now see an estimate of algorithmic involvement in their posts, notes, even comments. It's a narrow slice of functionality, but a visible one.
For Pangram, the integration represents something of a proof point. The company was founded in 2024 by engineers with backgrounds including work on Tesla Autopilot and Google AI projects, and it has positioned itself around a specific claim: fewer false positives. That matters because the alternative—flagging human prose as machine-generated—has become a credibility crisis for the field.
Consider the institutional retreat. The University of Waterloo pulled Turnitin's AI module in September 2024, citing accuracy concerns that administrators found untenable. Washington State followed in February. Then, this past June, the Los Angeles Times ran a piece documenting California students disputing AI-cheating charges. Turnitin, the incumbent plagiarism-detection giant, acknowledged what it called a "small but real false-positive risk." Small, perhaps. Real, definitely.
A Technical Pitch Built on Precision
Pangram's founders argue they have engineered around that risk. A November working paper from the University of Chicago's Becker Friedman Institute—one of the few independent evaluations available—found that Pangram achieved what the authors described as "a near zero FPR [false positive rate] and FNR [false negative rate]" in controlled testing. The study compared several detection tools; Pangram outperformed its peers on both metrics.
Nature weighed in with a July review of the broader AI-detection market, noting mixed reliability and troubling rates at which some tools misidentified human writing. The subtext: this is still an immature category.
Around the same time it announced funding, Pangram posted a technical report to arXiv for its fourth-generation model. The claims: better resistance to adversarial attacks—the tricks writers use to fool detectors—and improved generalization across domains. The company also previewed an image-detection capability, signaling ambitions beyond text. Whether that preview becomes a product remains to be seen.
Two Markets, One Pitch
Pangram lists 11 to 50 employees on LinkedIn, a range that suggests a still-lean operation. The business model straddles two audiences. On one side: academic integrity, with integrations into Canvas, Google Classroom, and other learning management systems. On the other: API-based content moderation for platforms trying to filter what some in the industry have started calling "AI slop."
That phrase—AI slop—has gained currency. Wired reported in April that researchers at Imperial College London, Stanford, and the Internet Archive used Pangram's tools in their work to quantify the problem online. Menlo Ventures' investment memo, published alongside the funding news, framed the bet in similar language: tackling "AI slop on the internet."
It's a subtle repositioning. Less classroom cop, more infrastructure play. Whether that framing helps as universities pull back from detection altogether is an open question.
What the Round Signals
The competitive landscape has shifted. GPTZero, a notable competitor in the space, was acquired by Superhuman in June—a deal that removed a well-funded player from the field. Turnitin, dominant in plagiarism detection for years, now faces policy blowback over its AI add-on.
Pangram says the $9 million will fund improvements to text-detection accuracy and scale its image work. The company runs on a subscription SaaS model with API access; pricing tiers have not been disclosed in detail, though a blog post from last November referenced a pricing adjustment.
One thing the round does not answer: whether enterprises and institutions will embrace detection at all, or whether the false-positive specter will keep buyers on the sidelines. Menlo's thesis is that authenticity infrastructure is inevitable. Pangram's challenge is proving the market agrees—and that its technology holds up under real-world pressure, not just in controlled studies.
For now, Substack users can click that button. What they do with the results is another story.
