TechCrunch: Pangram raises $9M and ships Pangram 4 with image detection

On July 29, 2026, Pangram Labs announced $9 million in funding led by Menlo Ventures and simultaneously released Pangram 4, a new version of its AI text detection model, alongside a research preview of an AI image detector.

The funding and product launches follow Pangram’s integration into Substack one week earlier. That integration let readers scan newsletter posts for AI-generated content estimates. Pangram 4 is the model behind that scanner and represents the company’s most accurate text detection release to date — achieving over 99% accuracy on internal benchmarks, with a stated false positive rate of 0.01%.

The more technically significant addition is what Pangram calls token-level attribution: the ability to identify which individual sentences within a document are likely AI-generated, rather than returning only a percentage estimate for the full text. This matters for publishers reviewing mixed human-AI drafts, where the concern is not whether AI was used at all but whether specific passages were generated and left unedited.

The AI image detector, released as a research preview and open to the public, claims 99.5% accuracy in internal testing. It works without relying on metadata or watermarking, analyzing pixel patterns instead — which means it is harder to defeat by simply stripping EXIF data or saving through a secondary export.

For editorial teams, the practical relevance of this release is less about individual detection queries and more about the infrastructure question it points toward. As AI-generated content becomes harder to distinguish from human writing, detection tools are likely to become standard audit components in editorial workflows, especially for publishers that receive unsolicited submissions or republish content from external contributors. Pangram’s Substack integration is the first large-scale deployment of this kind of transparency layer in a major publishing platform, and the product roadmap implied by this funding suggests similar integrations are in the pipeline for other platforms.