AI detection startup Pangram has raised $9 million and landed a partnership with Substack, but co-founder and CEO Max Spero says the hardest part of the job isn’t spotting fake content — it’s dealing with the gray area between human and machine authorship.
Spero joined TechCrunch’s Equity podcast to discuss the company’s approach, its new image detection tool, and why the industry’s framing of the problem as a simple “real or fake” binary is outdated. The interview comes as platforms from job boards to insurance claims adjusters grapple with a surge of AI-generated material that is increasingly difficult to distinguish from human output.
Also read: Enterprise AI contracts no longer guarantee revenue stability, Madrona research shows
Funding and growth amid rising demand for AI transparency
The $9 million round, which Pangram announced earlier this year, is earmarked for expanding its detection capabilities and scaling its partnerships. The Substack deal, unveiled in recent weeks, integrates Pangram’s technology directly into the publishing platform, allowing readers to see when authors have used AI assistance in their posts. It’s a notable step for a company that began with a focus on detecting AI-generated text in academic and professional settings.
Pangram’s pitch is that existing detection tools are too blunt. Spero argues that the real challenge is distinguishing between content that is fully AI-generated, lightly edited AI output, and human writing that uses AI as a brainstorming tool. “The line between AI-assisted and AI-generated is where the real trust issues lie,” he said on the podcast, emphasizing that his company’s models are designed to provide a more nuanced read on how AI was used in a piece of content.
Also read: Sequoia-backed Empirik raises $21M to predict IT outages before they happen
The startup’s expansion into image detection comes as visual AI tools like Midjourney and DALL-E become mainstream, complicating efforts by newsrooms and social platforms to verify photos. Pangram’s image tool aims to flag synthetic visuals with a similar probability-based approach, rather than a definitive label.
Why this matters for the broader internet trust problem
The stakes go beyond social media feeds. Recruiters are sifting through AI-generated resumes, e-commerce sites are battling fake reviews, and insurers are investigating AI-manufactured claim documentation. Each of these use cases demands a different threshold for what counts as unacceptable AI use, Spero noted. A job applicant might use AI to polish a cover letter, but an insurance claimant using AI to fabricate damage photos is fraud.
Pangram’s approach is to give platforms and enterprises the tools to set their own policies, rather than imposing a universal standard. That flexibility is a key selling point for partners like Substack, which must balance transparency with the creative freedom of its writers. The company is not alone in this space — competitors like GPTZero and Originality.ai have also raised significant funding — but Pangram’s focus on probabilistic scoring and enterprise partnerships distinguishes it from consumer-facing checkers.
The detection arms race is unlikely to end soon. As AI models improve, so too will the tools that identify them, creating a cat-and-mouse dynamic that Spero acknowledges. He told Equity that Pangram’s roadmap includes adapting to new AI architectures and expanding into audio and video detection, areas where synthetic media is growing rapidly.
For now, the Substack integration offers a real-world test of how AI transparency can work at scale. If successful, it could pave the way for similar features on other major platforms, giving readers more context about the content they consume. Whether that restores trust in the internet’s information ecosystem remains an open question, but Spero is clear that the industry must move beyond simplistic labels.
This article is for informational purposes only and does not constitute financial advice. The AI detection market is evolving rapidly, and any investment or business decisions should be made with appropriate due diligence.
