The New York detector behind Substack's reader-facing AI scan closed a $9 million round led by Menlo Ventures — and moved beyond text into AI image detection — as more platforms make "is this AI?" a built-in feature.
2026-07-29 · by Moe Ameen
Pangram, an AI content detection startup based in New York, raised a $9 million round led by Menlo Ventures, with participation from Haystack, ScOp, Script Capital, and Cadenza, as reported on July 29, 2026. The company was founded by Max Spero and Bradley Emi, two Stanford AI graduates, and had previously raised a seed round in 2025 before this larger raise. Treat the exact figures and investor list as a snapshot from the announcement and its coverage rather than audited numbers.
Pangram's pitch is accuracy on a hard problem. Its text detector estimates whether a passage is AI-generated or AI-assisted and highlights which segments read as AI, advertising accuracy above 99% and a false-positive rate of roughly 1 in 10,000 human documents — figures it says have been examined by outside researchers, including at the University of Chicago and the University of Maryland, alongside a top placement on the COLING 2025 benchmark. Rather than depend on watermarks or metadata, Spero has described a "synthetic mirror" method: the tool generates a matched AI version of your text and analyzes the stylistic gap, an approach that also lets it catch text laundered through AI "humanizer" paraphrasers.
Alongside the raise, Pangram pushed beyond text. Around July 29, 2026 it introduced an image detector as a research preview that analyzes pixel-level statistics to flag AI images across different generators — not just one provider's watermark — and can spot AI elements composited into an otherwise real photo. The company reaches users through a web app, a Chrome extension that labels AI-looking posts in feeds on X, LinkedIn, Substack, Reddit, and Medium, a Google Docs add-on, an API, and LMS integrations for schools.
The funding lands in a specific context: readers and platforms increasingly want a built-in way to ask whether content is AI. Substack integrated Pangram in July 2026 to power a reader-facing "Scan for AI text" check, and the direction of travel across the industry — platform AI labels, disclosure rules — points the same way. The money is a bet that detection becomes standard infrastructure, not a niche academic tool.
The useful way to act on this news is to stop treating detection as something to beat and start treating it as a proofing standard you build around — because the money says it isn't going away. Concretely, that means two moves, and Kompozy is built for both. First, govern the draft: Kompozy is a content generation and publishing engine, and when it writes a Blog Article, an Email Newsletter, or Text Posts, the Persona Brief holds your voice and a banned-word filter strips the generic AI-tell phrasing — the flat openers, the rule-of-three padding — that detectors and readers both react to. Then its per-post review pipeline keeps a human in the loop before anything ships, so you edit the copy into your own words and pair it with a clear "how I make this" disclosure. That human-reviewed, disclosed workflow is the opposite of feeding text through a "humanizer," which is precisely what Pangram now catches.
The second move is not to stake your presence on a single modality a detector reads. Pangram now scans images as well as text, and its extension labels feeds directly — so the resilient shape is a multi-format operation, not a text-only one. Kompozy generates Persona Shorts and HeyGen avatar video, Clipped Shorts, Carousels, Quote Graphics, and Photo Posts, then schedules and publishes across nine platforms from one queue, with you approving each piece. Run detection as an honest self-check on your writing; run Kompozy as the studio that produces a disclosed, human-in-the-loop, multi-format presence no single authenticity score can define.
Pangram, a New York AI content detection startup, raised a $9 million round led by Menlo Ventures, with participation from Haystack, ScOp, Script Capital, and Cadenza, reported on July 29, 2026. It also introduced an AI image detector as a research preview. Pangram's text detector advertises 99%+ accuracy and a roughly 1-in-10,000 false-positive rate and already powers Substack's "Scan for AI text" feature.
Its co-founder says it avoids watermarks and metadata. The text detector uses a "synthetic mirror" method — generating a same-topic, same-tone AI version of your text and analyzing the stylistic differences, trained on tens of millions of known-human documents — which also lets it catch text run through AI "humanizer" paraphrasers. The new image detector analyzes pixel-level statistics to flag AI images across different generators.
Detection is becoming a standard, reader-facing feature, so the safe posture is to disclose how you work and keep a genuine voice rather than try to evade a scanner — the "humanizer" tools built for that are exactly what Pangram catches. Practically, generate drafts with a governed brand voice, review and edit them yourself before publishing, and diversify beyond text. A content engine like Kompozy applies a Persona Brief and banned-word filter, keeps a human in the loop with a per-post review pipeline, and generates video, images, and carousels across nine platforms.