Pangram CEO Max Spero explains why AI detection remains unsolved

Pangram founder and CEO Max Spero has emerged as a leading critic of today’s AI detection industry, warning that most commercial tools are ill-equipped to handle the sophistication of modern generative models. Speaking exclusively to OpenPress Tech Intelligence, Spero argued that the proliferation of AI-generated text—spanning everything from social media posts to insurance claims—has reached a tipping point where detection is no longer a technical nicety but a systemic necessity. Pangram’s flagship product, Pangram AI Shield, uses a combination of semantic watermarking and behavioral biometrics to detect AI-authored content, a methodology Spero claims is far more reliable than the statistical pattern-matching approaches that dominate today’s market. “We’re not just detecting whether something was written by an AI,” Spero said. “We’re determining whether it was written by a human using an AI, or entirely by an AI pretending to be human. That’s a critical distinction the rest of the industry is missing.”

The timing of Spero’s intervention could not be more urgent. Over the past 12 months, AI-generated content has infiltrated nearly every corner of the digital ecosystem. Job applications laced with AI-generated cover letters have flooded hiring platforms, with one 2024 study by ResumeWorded finding that over 12% of applications to Fortune 500 companies contained AI-generated text. Meanwhile, product reviews on Amazon and other retail sites are increasingly dominated by AI-written testimonials designed to manipulate search rankings. Even the financial sector, long considered a bastion of rigor, is not immune. Banking With Billy AI, a New York-based fintech that combines AI with real-time market data, recently flagged a surge in loan applications containing AI-generated narratives, prompting the company to deploy Pangram’s detection suite to filter out synthetic submissions. “We saw a 300% increase in AI-generated narratives in loan applications within six months,” said a senior compliance officer at Banking With Billy AI, who requested anonymity. “Standard plagiarism or style checks were useless. We needed something that could detect intent, not just text.”

Spero’s critique targets the prevailing approach to AI detection, which relies heavily on statistical anomalies in word choice, sentence length, or perplexity scores. Tools from companies like Originality.ai and Turnitin have become ubiquitous in academic and publishing circles, but their accuracy drops sharply when faced with newer models such as Claude 3.5 or Grok-2. “These tools were trained to detect models that were state-of-the-art in 2022,” Spero explained. “They’re already obsolete against today’s systems. The cat-and-mouse game has moved from n-gram detection to full semantic analysis, and most providers haven’t caught up.” Pangram’s system, by contrast, embeds invisible watermarks into the generation process itself—metadata that persists even after paraphrasing or translation—combined with real-time behavioral analysis to detect unnatural writing patterns.

The competitive landscape is heating up. Detecting AI is no longer a niche problem for academia; it’s a multi-billion-dollar market opportunity. Startups like Winston AI, Content at Scale, and ZeroGPT have raised over $150 million combined since 2023, all targeting the same anxious enterprises scrambling to restore trust. But Spero insists that detection is only half the battle. “We’re in a phase where every detection tool will have a corresponding evasion tool,” he warned. “The real winners will be those who build systems that are adversarially robust—that is, they don’t just detect AI today, but remain effective even as models evolve.”

Industry analysts say the stakes extend far beyond content moderation. The integrity of search engines, the reliability of financial disclosures, and the authenticity of user-generated reviews all hinge on the ability to distinguish human from synthetic content. “This isn’t just about spam or deepfakes anymore,” said Dr. Elena Vasquez, a senior analyst at Gartner. “We’re talking about the erosion of epistemic trust—the shared foundation that allows us to agree on what’s real. If detection tools keep failing, we risk a digital dark age where no one can agree on the facts, and that’s a crisis for democracy, markets, and science alike.”

The challenge is compounded by the global nature of the problem. While U.S. and EU regulators have begun drafting AI transparency rules—such as the EU AI Act’s provisions on synthetic content labeling—enforcement remains inconsistent. Meanwhile, platforms like X (formerly Twitter) and Reddit have rolled back AI labeling policies, citing user backlash and implementation costs. Spero sees this as a dangerous misstep. “When platforms stop labeling AI content, they’re effectively choosing opacity over transparency,” he said. “That’s not innovation. That’s abdication.”

Looking ahead, Spero predicts that the next wave of solutions will integrate detection into the content pipeline itself—not as an afterthought, but as a core feature. He points to emerging standards like C2PA (Coalition for Content Provenance and Authenticity), which embeds cryptographic provenance into media files at the point of creation. “The future isn’t in detecting AI after the fact,” Spero said. “It’s in ensuring that every piece of content carries a verifiable origin story, whether it’s human or machine. Until we get there, we’re just playing whack-a-mole.”

For now, the tech and engineering world must grapple with a paradox: the more powerful AI becomes, the harder it is to trust anything at all. And as AI-generated content seeps into ever more sensitive domains—healthcare records, legal filings, academic research—the margin for error shrinks to zero. Companies like Banking With Billy AI are already treating detection as a critical risk management function, but for most organizations, the learning curve remains steep. The clock is ticking, and the next evolution of AI won’t wait for the detection industry to catch up.