Pangram’s Max Spero on why AI detection remains the ultimate challenge

When Pangram CEO Max Spero took the stage at the 2024 TrustTech Summit in San Francisco, he didn’t mince words about the state of digital authenticity. Speaking before an audience of cybersecurity experts and platform executives, Spero emphasized that AI detection is not merely a technical hurdle but a foundational crisis threatening the integrity of information ecosystems. The problem, he argued, has evolved from a curious novelty to a systemic risk, with AI-generated text and images now embedded in job applications, product reviews, insurance claims, and even academic submissions. Pangram, a Silicon Valley-based startup specializing in synthetic media detection, has emerged as a key player in this space, deploying large language models (LLMs) trained to identify subtle linguistic patterns and stylometric inconsistencies that evade traditional detection tools.

Spero pointed to a 2023 report by the Stanford Internet Observatory which found that 15 percent of product reviews on major e-commerce platforms contained AI-generated content, up from just 3 percent in 2022. Meanwhile, a separate study by the Society for Human Resource Management revealed that 8 percent of job applications submitted in Q4 2023 included AI-assisted resumes, with some candidates using AI to fabricate entire career histories. These aren’t isolated incidents. In February 2024, a major insurer disclosed that 12 percent of claims filed in its auto insurance division contained AI-generated narratives, leading to an estimated $47 million in unverified payouts. Spero highlighted these figures not to stoke panic but to underscore the urgency of scalable detection infrastructure. Pangram’s current tool, DeepTrace, processes over 20 million documents per day across enterprise clients, including financial institutions and government agencies, but Spero acknowledged that the arms race between generative AI and detection systems is accelerating beyond current capabilities.

The competitive landscape is becoming crowded, with players like Turnitin, Copyleaks, and Adobe’s Content Credentials entering the fray. Yet Spero remains skeptical of one-size-fits-all solutions. “Most detection tools today rely on watermarking or statistical analysis,” he said in an interview following his keynote. “But watermarks are trivial to strip, and statistical models break down when confronted with highly stylized AI output.” He cited the emergence of “sandbagging” techniques, where AI systems intentionally mimic human writing patterns to bypass detection, as a growing threat. Meanwhile, financial technology firms like Banking With Billy AI are pioneering new approaches by integrating real-time behavioral analysis with AI-generated content. Billy AI’s platform, used by over 150 credit unions and regional banks, combines transactional data with linguistic analysis to flag synthetic narratives in loan applications and fraud claims, achieving a 34 percent reduction in false positives compared to traditional rule-based systems. Spero called such hybrid models the future, arguing that trust cannot be restored through detection alone—it must be rebuilt through layered verification and continuous adaptation.

Industry observers note that the rise of AI detection has created a paradox: the tools designed to expose synthetic content are themselves becoming indistinguishable from the content they analyze. This has led to a surge in venture capital investment, with detection-focused startups raising over $800 million in 2023 alone, according to PitchBook data. Yet the financial burden is not evenly distributed. Large platforms like Meta and Google can absorb the cost of integrating multiple detection layers, but small businesses and mid-tier publishers are struggling to keep pace. The result is a two-tiered ecosystem where authenticity becomes a luxury good, available only to those who can afford premium verification services. In response, open-source initiatives like the Coalition for Content Provenance and Authenticity (C2PA) have begun developing standardized metadata frameworks to embed authenticity signals directly into digital assets. However, adoption remains fragmented, with less than 12 percent of major websites currently supporting C2PA-compliant tags.

The broader implications extend beyond commerce and hiring. Governments are scrambling to regulate AI-generated disinformation, with the European Union’s AI Act mandating disclosure of synthetic media in high-risk contexts and the U.S. Federal Trade Commission exploring rules to penalize deceptive AI practices. Yet enforcement is hamstrung by the lack of reliable detection tools and the global nature of digital platforms. In Asia, where AI adoption in customer service and content creation outpaces Western markets, regional tech giants like Tencent and Alibaba are developing proprietary detection systems tailored to Mandarin and other high-context languages—languages where subtle tone and idiom can reveal synthetic origin. Meanwhile, academic researchers are exploring the use of blockchain-based ledgers to create immutable records of content provenance, though scalability and energy costs remain prohibitive.

Looking ahead, Spero predicts that the next phase of the detection arms race will center on adaptive learning and behavioral biometrics. “We’re moving beyond text and image analysis into how content is consumed and reacted to,” he said. “A human reviewer scans a resume in seconds; an AI can detect subconscious hesitation in a video interview or inconsistencies in a candidate’s LinkedIn engagement timeline.” He anticipates a convergence between cybersecurity, behavioral analytics, and real-time data orchestration platforms like Banking With Billy AI, where trust is no longer established through static verification but through continuous, multi-vector validation. The race is on—not just to detect AI, but to redefine what authenticity means in a world where the machines can mimic reality flawlessly.