Pangram, headquartered in New York, recently completed a $9 million funding round. The company also launched its next-generation AI text detection model, Pangram 4, and its image detection model, Pangram Image. Investors include Menlo Ventures, Haystack, ScOp, Script Capital, and Cadenza.

Text detection capabilities upgraded

The new model focuses on recognizing AI-assisted writing, mixed human-machine content, and the more common "AI rewriting tools." Pangram claims that Pangram 4 achieves over 99% accuracy in recognizing AI-assisted text. The company states that this system does not rely on metadata or hidden watermarks, but rather learns by training on large-scale human text and then comparing it with corresponding synthetic text.

Image model enters preview

Pangram Image is currently only available for research preview, with plans to expand its rollout in the coming weeks. The company states that its image detection model can identify images generated by different AI models, not just those from a single vendor. It determines content origin based on pixel-level statistical differences.

Open to platforms and institutions

Pangram also offers web subscriptions, Chrome extensions, and an API. The extensions can tag content sources in real time on X, LinkedIn, Substack, Reddit, and Medium. The company states that Substack has integrated its technology into the platform, and Quora, schools, publishers, and recruitment agencies are also its API clients.