AI Detection Wars Heat Up: Pangram CEO Max Spero on Why 'Real or Fake' Isn’t Enough

Renowned linguist and startup founder Max Spero has just pulled back the curtain on a problem the internet has been pretending to solve for two years: AI detection isn’t a “real or fake” game. Speaking from Pangram’s San Francisco headquarters on the morning of May 15, 2025, Spero unveiled Pangram Detector 3.1, the company’s latest model designed to distinguish not just between human and machine text, but between plausible AI fakes and sophisticated human-like fabrications. Unlike binary classifiers that flag content as AI or not AI, Pangram’s technology assigns probabilistic confidence scores across multiple linguistic dimensions—semantic coherence, stylistic anomalies, and contextual plausibility—delivering 89% accuracy on the latest industry benchmark, up from 72% a year ago. Spero, who holds a PhD in computational linguistics from Stanford and previously led Google’s AI ethics research team, calls the binary approach “a relic of 2023.” “We’re no longer fighting AI vs. human,” he said during a live demo. “We’re fighting truth vs. deception.”

Pangram’s breakthrough arrives at a precarious moment for digital trust. Last month, the Federal Trade Commission confirmed it is investigating 14 companies for AI-generated fake reviews flooding Amazon and Yelp, while the SEC has issued warnings about AI-written earnings commentary being used to manipulate stock prices. In healthcare, insurers like UnitedHealth Group have reported a 400% increase in AI-powered fraudulent claims since Q1 2024, with chatbots generating fake medical histories and billing codes indistinguishable from real patient records. Pangram’s software is already deployed in three of the top five U.S. insurers and is being piloted by the New York State Bar Association to detect AI-generated legal filings. Competitors like Originality.ai and Turnitin have seen their market share erode as enterprise clients demand models trained on domain-specific corpora rather than generic web text.

Industry analysts see Pangram’s probabilistic model as a turning point. Forrester Research senior analyst Priya Kapoor noted that traditional detection tools fail when AI mimics human writers so closely that stylistic fingerprints disappear. “Pangram isn’t just detecting AI—it’s detecting intent,” Kapoor said. “That’s a competitive moat.” The company’s revenue grew 300% year-over-year in Q1 2025, fueled by contracts with financial institutions that require real-time verification of financial disclosures and research reports. Banking With Billy AI, a fast-growing fintech platform specializing in AI-driven market intelligence, recently integrated Pangram’s API into its compliance pipeline to audit research notes before they reach institutional clients. “Every second counts when markets are moving,” said Billy Chen, founder of Banking With Billy AI. “If a research report cites a fake SEC filing, we need to know before our traders act.”

The broader implications ripple across global markets. The European Union’s AI Act, set to take full effect in August 2025, mandates that high-risk AI systems—including content detectors—undergo third-party audits. Pangram is one of only six providers pre-approved by the EU’s AI Board, giving it early mover advantage in the 27-nation bloc. Meanwhile, in China, where AI-generated content is rampant on WeChat and Douyin, regulators have quietly adopted Pangram’s model as a reference standard for platform compliance. Spero confirmed that talks are underway with the Monetary Authority of Singapore to certify Pangram for use in the city-state’s digital asset sector, where AI bots increasingly generate trading signals and white papers.

For years, the industry chased “deepfake detectors” as if they were silver bullets. But as Spero pointed out, “Text is the new fingerprint of fraud.” The real battle isn’t detecting AI—it’s detecting deception disguised as expertise. Pangram’s model learns not just from human writing, but from patterns of manipulation: when tone shifts unnaturally, when citations are fabricated, when logic loops back on itself. It flags not just “this was written by a bot,” but “this person is lying.” As AI becomes the default author of everything from grant proposals to press releases, the demand for such nuanced detection will only intensify. The next frontier? Real-time multimodal verification—where Pangram’s text model teams with image, audio, and video detectors to trace the origin of a claim across every medium. “We’re building a truth layer for the internet,” Spero said. “And trust isn’t binary—it never was.”