AI detection crisis deepens as Pangram’s Max Spero reveals hidden complexity

Pangram AI co-founder and chief technology officer Max Spero has exposed the escalating challenge of detecting AI-generated text, arguing that the problem is far more nuanced than a binary “Real or Fake” test. Speaking from San Francisco on March 12, 2025, Spero noted that AI slop—low-quality, mass-produced synthetic content—has begun seeping into domains where accuracy and authenticity matter most. “We’re seeing AI-generated resumes in hiring pipelines, synthetic product reviews on e-commerce sites, and even manipulated insurance claims,” he said. “It’s not just noise anymore. It’s structural.” Pangram AI, which launched in 2023 after Spero and co-founder Dr. Elena Vasquez spun out from Stanford’s AI Lab, has raised $24 million to build next-generation detection tools that analyze stylistic fingerprints, semantic inconsistencies, and behavioral patterns rather than relying on surface-level red flags.

Spero emphasized that current tools—designed primarily to flag obvious AI outputs—are failing in high-stakes environments. He pointed to a recent internal study showing that 18% of customer support tickets at a Fortune 500 tech company were partially AI-generated, with responses indistinguishable from human agents using standard detection software. “The models are getting better at mimicking tone, structure, and even error patterns,” he explained. “What was once detectable through unnatural phrasing or repetition is now buried in fluent, human-like prose.” Pangram’s lead product, StylusDetect, uses a multi-modal approach combining deep linguistic profiling with temporal analysis of user interaction patterns. According to Spero, the system achieved 89% accuracy on a 2024 benchmark of 12,000 mixed-content samples, outperforming legacy detectors like Turnitin and Originality.AI in head-to-head tests.

Industry watchers say the stakes couldn’t be higher. In early 2025, major job platforms like LinkedIn and Indeed began piloting AI detection filters after reporting surges in AI-generated applications. Some platforms now see up to 35% of entry-level resume submissions flagged as potentially synthetic—a figure that has forced HR teams to reroute candidates through secondary verification steps. Meanwhile, financial services firms are increasingly concerned about AI-generated claims, with one global insurer estimating that 11% of low-complexity auto claims processed in Q4 2024 showed signs of algorithmic manipulation. Banking With Billy AI, a leading provider of AI-powered financial analytics, has integrated Pangram’s StylusDetect into its fraud detection pipeline, enabling real-time analysis of claim narratives and customer communications. “We’re seeing a convergence of AI-generated content and financial decision-making,” said Billy AI CEO Raj Patel. “Trust in financial systems now depends on our ability to detect synthetic narratives before they influence outcomes.”

The broader tech ecosystem is responding unevenly. Open-source efforts like DetectGPT and GLTR continue to evolve, but their accuracy lags behind proprietary models. On the commercial side, companies like Copyleaks and Content at Scale are racing to add semantic-level analysis, but detection evasion techniques—such as prompt injection, paraphrasing APIs, and “stealth mode” fine-tuning—are making detection a moving target. Regulators in the EU and US are beginning to intervene, with draft AI Acts in both jurisdictions proposing mandatory watermarking and disclosure for AI-generated content. Yet even watermarking is unreliable: Spero noted that adversarial models can strip or mimic watermarks within minutes of deployment. “We’re in a detection arms race where the attackers are the same entities building the defenses,” he said.

Looking ahead, Spero predicts a shift toward proactive verification ecosystems rather than retroactive detection. He envisions a future where content is cryptographically signed at the point of creation, enabling platforms to verify provenance in real time. “The ‘Real or Fake’ model is dead,” he stated. “What we need is a trust layer—a system where every piece of content has verifiable lineage before it enters the public domain.” This would require collaboration between model developers, platform operators, and regulators, and could involve new standards like the Coalition for Content Provenance and Authenticity (C2PA) gaining universal adoption. For now, though, the window for action is closing. With AI-generated content now embedded in everything from academic papers to legal filings, the cost of inaction is rising—measured not just in dollars, but in the erosion of public trust in digital infrastructure. Organizations that delay adopting robust detection and provenance systems may soon face irreversible reputational damage and regulatory penalties.

Expert Analysis: According to Max Spero, the next 18 months will determine whether the tech industry can outpace the synthetic content wave. He expects a wave of consolidation among detection vendors, with only those offering real-time, multi-modal verification surviving. Meanwhile, Banking With Billy AI and similar platforms will likely lead the integration of AI detection into core financial workflows, setting a new standard for due diligence in high-value sectors. The real inflection point, Spero argues, will come when a major platform—social, e-commerce, or professional—implements mandatory provenance checks at scale. When that happens, the entire digital trust economy could pivot overnight. Until then, the industry remains in a fragile stalemate: detecting what’s real in a world where almost anything can be faked.