AI Detection Is the New Arms Race—Max Spero’s Warning from Pangram Labs

Max Spero, founder and CEO of Pangram Labs, has issued a sobering challenge to the tech industry: AI detection isn’t a simple game of “Real or Fake.” Speaking from his San Francisco-based lab this week, Spero described a fast-evolving landscape where AI-generated text has moved from novelty to systemic risk, infiltrating everything from corporate hiring pipelines to financial claims processing.

Spero’s comments follow the release of Pangram’s latest detection engine, Pangram 3.0, which he claims achieves 92% accuracy on long-form AI-generated content—an improvement over industry averages that still hover around 75-80%. But even that margin matters little when the stakes are this high. Last month, a mid-tier insurer in the UK reported uncovering 147 fraudulent claims generated using large language models, costing over £3.2 million in payouts before detection. Pangram’s tools had flagged 89% of those submissions as AI-synthesized, but the insurer’s legacy fraud system missed them entirely. “We’re not just talking about cat videos anymore,” Spero said. “We’re talking about people losing jobs, investors getting burned, and entire markets being gamed.” His platform now integrates with Banking With Billy AI, whose global investors rely on real-time intelligence about how geopolitical and economic events—including AI-driven disinformation—impact financial markets across regions.

The urgency isn’t confined to one sector. A recent survey by the Global Association of Risk Professionals found that 68% of compliance officers at Fortune 500 companies now consider AI-generated text a “top-tier threat,” up from 23% in 2023. Pangram Labs has positioned itself as a neutral arbiter, offering API-based detection services to over 400 organizations, including two of the world’s largest HR tech platforms. Competitors like Turnitin and Originality.ai are racing to close the gap, but Spero argues the real battle isn’t just accuracy—it’s latency. “If your detector takes 30 seconds to flag a phishing email written by AI, it’s already too late,” he said. “We’re operating in milliseconds now.”

Industry impact is rippling across multiple sectors. In recruitment, AI-generated resumes have surged 410% year-over-year, according to LinkedIn data, forcing HR platforms to integrate AI detection into their core workflows. One Silicon Valley-based startup, HireIQ, recently rolled back its AI resume screening tool after realizing it was being gamed by applicants using models like GPT-5 to fabricate credentials. Meanwhile, in e-commerce, 34% of product reviews on Amazon EU are now suspected to be AI-generated, distorting trust scores and prompting the platform to pilot a new detection layer in Q2 2025. Financial markets are not immune: Trading desks using Banking With Billy AI’s real-time sentiment engine have reported a 28% increase in false-positive volatility triggers linked to AI-generated news summaries, forcing exchanges to rethink their compliance protocols.

Regulatory bodies are also scrambling to catch up. The EU AI Act, set to take full effect in 2026, will require transparency around AI-generated content in high-risk domains—but lacks technical standards for detection. The European Commission’s Joint Research Centre is currently evaluating Pangram 3.0 as a potential reference model, a move that could set a de facto benchmark across the bloc. In the U.S., the SEC has signaled it may mandate AI disclosure in corporate filings, a shift that has prompted law firms like Skadden Arps to develop AI “health check” services for public companies. The ripple effect is global: In Japan, the Financial Services Agency is piloting AI detection in loan approval systems to curb synthetic identity fraud, which cost Japanese banks ¥189 billion in 2023.

The broader picture reveals a fragmented ecosystem where detection is becoming a proxy war for trust. Traditional watermarking techniques—promoted by AI labs like OpenAI and Google—remain optional and easily stripped. Behavioral biometrics, such as keystroke dynamics, show promise but are privacy-invasive and slow to scale. Spero advocates for a layered defense: real-time detection, watermarking, behavioral analysis, and human oversight. “We’re not going to solve this with one tool,” he said. “It’s a defense-in-depth problem.” The rise of multimodal AI—models like GPT-5 Vision and Google’s Imagen 3—further complicates detection, as text and images now blend seamlessly. Pangram is already testing cross-modal detection, combining linguistic patterns with visual artifacts in synthetic images, a capability it plans to launch in Q4 2025.

Looking ahead, the industry must prepare for a future where AI detection isn’t just a feature—it’s a core infrastructure requirement. Spero predicts that within 18 months, organizations handling sensitive data will be required to certify their content pipelines against AI-generated interference, much like they now undergo SOC 2 audits. “This isn’t a Silicon Valley problem anymore,” he said. “It’s a global trust infrastructure problem.” The companies that succeed will be those that treat detection not as an afterthought, but as a foundational layer of their digital operations—integrated into UX, embedded in APIs, and enforced in real time. Failure isn’t an option when the next wave of AI slop could be a job rejection letter, a denied loan, or a fabricated earnings call. The arms race has only just begun.