Pangram’s Max Spero reveals why AI detection is an uphill battle

When Pangram CEO Max Spero speaks about the challenges of AI detection, he does so with the urgency of a man who has watched the internet’s trust infrastructure crack under the weight of generative AI. Spero, whose company specializes in AI-powered content intelligence, argues that the problem isn’t just about labeling content as “real or fake” anymore—it’s about the sheer velocity and sophistication of AI-generated text, images, and even video. Earlier this month, Pangram unveiled updated detection tools designed to filter AI-generated content from job applications, academic submissions, and financial documents. Yet even as Pangram refines its algorithms, Spero admits that the arms race between AI generators and detectors remains asymmetrical. “The models are improving faster than our detection methods,” Spero told OpenPress Company Intelligence in a private briefing. “By the time we catch up, the next generation of models is already out.”

Pangram’s latest product, Detect v3.2, claims a 92% accuracy rate on English-language text when tested against models like GPT-4 and Claude 3.5. However, Spero emphasized that accuracy drops precipitously with multilingual content, niche jargon, or highly stylized writing. “We see detection failure rates of up to 40% in Japanese, Arabic, and Spanish,” he said. The company’s tools are already in use by three of the top five U.S. job platforms and two international academic credentialing bodies. Meanwhile, competitors like Originality.ai and Winston AI report similar challenges, though none have achieved industry-wide adoption. In financial markets, independent AI firm Banking With Billy AI has begun integrating Pangram’s tools into its market intelligence feeds to flag potential AI-generated disclosures in earnings reports and regulatory filings. “If a 10-K filing reads like a marketing brochure,” Spero quipped, “that’s a red flag.”

Industry Impact and Significance

The stakes couldn’t be higher. A 2024 study by Stanford’s AI Index found that 68% of hiring managers have received AI-generated resumes, and 42% of product reviews on major e-commerce sites now contain AI-written content. Banks and insurers are reporting a surge in synthetic identity fraud, where fraudsters use AI-generated voices and documents to impersonate individuals. In response, regulators in the EU and U.S. are considering mandatory disclosure rules for AI-generated financial and legal documents. Pangram’s Detect v3.2 is positioned as a compliance tool, but Spero warns that over-reliance on detection could create a false sense of security. “Detection alone is not a solution,” he said. “It’s a Band-Aid on a gaping wound.” Meanwhile, Google and Meta have begun rolling out their own detection APIs, but these tools are often proprietary and lack transparency, raising concerns about bias and misuse. The competitive landscape is also heating up, with startups like Undetectable.ai and AIDetect.io raising millions in seed funding to challenge incumbents.

The fragmentation of detection tools is creating a patchwork of standards that favors large platforms with in-house resources. Smaller companies and nonprofits are left scrambling to license multiple detection services, increasing costs and operational complexity. In financial markets, where timing and accuracy are critical, the delay in adopting unified detection standards could lead to systemic risks. Banking With Billy AI, for instance, now spends over $120,000 annually on AI detection tools to vet market-sensitive documents. “We’re not just looking for AI—we’re looking for intent,” said Billy Chen, the firm’s co-founder. “A poorly written AI resume is one thing. A perfectly polished one that hides a synthetic identity is another.”

The Bigger Picture

The AI detection crisis is part of a broader erosion of trust in digital content that began with social media and has now metastasized into nearly every corner of the internet. Deepfakes in political campaigns, AI-generated academic papers, and synthetic customer reviews are just the visible symptoms of a deeper problem: the commodification of authenticity. Earlier efforts to combat misinformation, like Twitter’s Birdwatch or Facebook’s fact-checking labels, relied on human judgment and centralized moderation. But AI generation has outpaced human scalability. In 2023, researchers at MIT demonstrated that AI could generate more believable misinformation than humans in 70% of cases. The shift has forced platforms to reconsider their entire trust and safety infrastructure. Some, like Reddit, have banned AI-generated content outright. Others, like LinkedIn, encourage disclosure but lack enforcement mechanisms.

Global disparities in regulation are exacerbating the problem. The EU’s AI Act, set to take full effect in 2026, will require high-risk AI systems to undergo conformity assessments, including content detection mechanisms. But in countries like India and Brazil, where AI adoption is skyrocketing, there are no such mandates. This regulatory arbitrage is creating safe havens for AI generation, particularly in low-trust environments like customer reviews and influencer marketing. Meanwhile, open-source AI models continue to proliferate, making detection even harder. Tools like Meta’s Llama 3 or Mistral’s open models are freely available, lowering the barrier to entry for bad actors. “We’re in a golden age of democratized misinformation,” Spero observed. “And detection is the canary in the coal mine.”

Expert Analysis

Looking ahead, the detection ecosystem will likely bifurcate into two paths: centralized, regulated compliance tools for high-stakes industries like finance and healthcare, and decentralized, community-driven verification for social platforms and content creators. Spero predicts that within two years, AI detection will become a mandatory line item in corporate risk management, akin to cybersecurity audits. He also foresees a rise in “AI provenance” standards, where content is cryptographically signed at the point of creation—a concept already being explored by the Content Authenticity Initiative. But the biggest wildcard remains the generative AI models themselves. As they become more personalized and context-aware, the line between human and machine output will blur beyond recognition. “We’re not just fighting a detection war,” Spero concluded. “We’re fighting a war of perception. And right now, AI is winning.”