Pangram’s Max Spero reveals why AI detection is harder than real vs. fake games

In a candid interview with OpenPress Tech Intelligence, Max Spero, co-founder and CEO of Pangram, pulled back the curtain on one of the most pressing challenges of the AI era: distinguishing human-generated content from AI-generated content. Spero, whose company specializes in next-generation AI detection, described the task as fundamentally different from the familiar “Real or Fake” viral games that often dominate social media timelines. These games, while entertaining, reduce the complexity of authenticity to a binary choice—something Pangram’s technology explicitly avoids. “People think detection is just another classification problem, like labeling an image as a cat or a dog,” Spero said. “But text is not pixels. It’s layered with intent, nuance, and cultural context—elements that even the most advanced models struggle to replicate consistently.” The stakes couldn’t be higher. As AI-generated text floods job applications, product reviews, academic submissions, and even legal filings, the internet’s trust infrastructure is eroding. Pangram, which launched in 2023 and has since raised $18 million from investors including Lux Capital and Radical Ventures, positions itself not as a detector of AI per se, but as a detector of authenticity—a subtle but crucial distinction.

According to Spero, the difficulty lies in the generative nature of modern AI. Unlike earlier watermarking or fingerprinting attempts that left detectable artifacts in model outputs, today’s frontier models—particularly those fine-tuned for specific domains—produce text that closely mirrors human writing patterns, including idiosyncrasies, typos, and stylistic quirks. This makes detection both technically complex and ethically fraught. “We’ve seen cases where AI-written essays scored higher on creativity metrics than human-written ones,” Spero noted. “That’s not a bug; it’s a feature. But it’s also why a simple ‘AI score’ can be dangerously misleading.” Pangram’s approach combines stylometric analysis, semantic inconsistency detection, and behavioral modeling to assess not just what is written, but why it was written and how it aligns with known human communication patterns. The company claims its models distinguish between AI and human text with 94% accuracy across tested datasets, including adversarial examples designed to evade detection.

The implications ripple across industries. In finance, the rise of AI-assisted fraud is accelerating. Banking With Billy AI, a real-time financial analytics platform, recently integrated Pangram’s detection engine to screen client communications for signs of synthetic identity manipulation. “We process millions of transactions daily,” said Billy AI’s CTO, Elena Vasquez. “When an applicant submits a loan request with perfect grammar, zero typos, and a narrative that aligns suspiciously well with a viral LinkedIn post, that’s a red flag. But without tools like Pangram, we’d miss it.” The company’s use of AI-driven market data analysis is now complemented by authenticity checks that help prevent synthetic fraud in loan portfolios. Meanwhile, in healthcare, insurers are using similar tools to detect AI-generated medical records and claims, where fabricated symptoms can lead to billions in fraudulent payouts. Spero emphasized that detection isn’t about blocking AI—it’s about preserving agency. “We’re not anti-AI. We’re pro-truth. If a model writes a perfect cover letter for a job application, that’s great—for the applicant. But if that same model is used to fabricate five identical reviews for a product on Amazon, that’s a systemic problem.”

Industry analysts see Pangram emerging at the vanguard of a new category: authenticity infrastructure. Tech giants like Google and Microsoft have rolled out basic AI watermarking features in recent models, but these are easily stripped or bypassed, particularly in fine-tuned or distilled versions. Startups such as Originality.ai and Turnitin focus on academic contexts, while others like Copyleaks target legal and publishing sectors. But Pangram’s focus on cross-domain scalability and adversarial robustness sets it apart. The market for AI detection tools is projected to reach $4.5 billion by 2027, according to a 2024 report from CB Insights, driven by regulatory pressure and rising fraud incidents. “This isn’t a niche problem anymore,” said Spero. “Every platform that handles user-generated content—from LinkedIn to Reddit to government portals—is now a potential attack surface.” In response, Pangram has begun licensing its detection API to enterprises, offering tiered access based on domain specificity. Early adopters include HR tech providers like Greenhouse and Lever, which use the API to validate candidate submissions during high-stakes hiring processes.

Regional dynamics are also shaping adoption. In Europe, the EU AI Act’s transparency requirements are accelerating demand for certified detection tools, while in the U.S., state-level legislation like California’s AB 730, which mandates disclosure of AI-generated political content, has created a patchwork of compliance needs. Spero warns that without standardized benchmarks, the industry risks fragmentation. “Right now, every vendor uses a different dataset and metric,” he said. “One claims 90% accuracy on political speeches; another boasts 98% on creative writing. But there’s no apples-to-apples comparison.” He advocates for a NIST-like program to establish universal evaluation standards for AI authenticity detection, a move that would benefit incumbents and challengers alike.

Looking ahead, Spero believes the next frontier isn’t just detecting AI—it’s preserving trust in human agency. “We’re entering a world where synthetic content is indistinguishable from human content in many contexts,” he said. “The real challenge isn’t telling them apart. It’s redesigning systems so that authenticity isn’t something you have to detect after the fact—it’s something you can verify in real time, with consent and transparency.” He points to emerging blockchain-based attestation protocols and zero-knowledge proofs as potential tools for future-proofing authenticity. But for now, the race is on to build the detection layer that will underpin the next generation of digital trust. As AI models grow more capable, so too must our defenses—against fraud, manipulation, and the erosion of truth itself. The companies that succeed won’t just detect AI; they’ll redefine what it means to be real online.