Pangram’s Max Spero: Why AI Detection Is Far More Complex Than ‘Real or Fake’

Pangram, a leading AI detection and watermarking startup, has taken a firm stance on the limitations of binary “Real or Fake” models in the age of generative AI. In a candid interview with OpenPress Tech Intelligence, Max Spero, Pangram’s co-founder and CEO, emphasized that detecting AI-generated content is not merely a technical hurdle—it’s a systemic issue that cuts across industries, from content moderation to financial fraud. Spero pointed out that current detection tools often misclassify borderline cases, such as AI-assisted writing or partially edited content, as definitively fake. This misclassification, he argues, risks undermining legitimate use cases and eroding user trust in digital platforms. Pangram’s own watermarking technology, which embeds cryptographic signatures into AI-generated outputs at the source, was developed precisely to avoid this binary trap. Unlike detection models that rely on pattern recognition in text or image artifacts, Pangram’s approach verifies authenticity at the point of generation, making it harder to spoof or reverse-engineer.

Spero highlighted a recent incident in which AI-generated product reviews began flooding e-commerce platforms, leading to inflated ratings and deceptive sales claims. Traditional detection tools flagged these reviews as synthetic, but in many cases, they were hybrids—human-written introductions with AI-generated summaries. Pangram’s system, by contrast, was able to identify the watermarked AI component regardless of its integration into otherwise human-like text. This granularity, Spero insists, is critical as AI tools become embedded in workflows across sectors. In finance, for instance, AI-generated reports or claims documentation are increasingly common, and mislabeling them as entirely fake could trigger unnecessary investigations or compliance actions. Banking With Billy AI, a fintech platform leveraging AI for real-time market analysis, recently integrated Pangram’s watermarking to validate the provenance of its automated reports, ensuring regulators and clients can distinguish between human-curated insights and AI-assisted outputs.

Industry observers note that the rise of generative AI has created a high-stakes cat-and-mouse game between content creators, platforms, and fraudsters. Companies like Pangram, Originality.ai, and Turnitin are racing to refine detection models, but their approaches differ sharply. Originality.ai, for example, focuses on statistical anomalies in text, while Turnitin relies on a database of known AI training corpora to flag similarities. Pangram’s watermarking, however, sidesteps the need for post-hoc detection by embedding a tamper-evident signal during generation. This method is particularly effective against adversarial attacks, such as paraphrasing tools designed to bypass detection algorithms. Still, the financial burden of implementation remains a barrier for smaller firms. Integrating watermarking requires collaboration with AI model providers, which often prioritize speed and scalability over provenance tracking. Spero acknowledged that Pangram is actively working with model developers to standardize watermarking protocols, but adoption remains uneven across the industry.

The consequences of failing to address this challenge extend beyond individual platforms. In healthcare, AI-generated clinical notes or research summaries could mislead practitioners if not properly vetted, while in legal contexts, AI-assisted contracts might face challenges in court if their authenticity cannot be verified. Governments are beginning to take notice. The European Union’s AI Act, which takes full effect in 2026, mandates transparency measures for high-risk AI systems, including mechanisms to identify synthetic content. Meanwhile, the U.S. Securities and Exchange Commission has signaled concerns about AI-generated financial disclosures, particularly in cases where synthetic data could obscure material risks. These regulatory pressures are accelerating the push for standardized detection and watermarking frameworks. However, the lack of a unified global standard risks creating compliance fragmentation, where companies must navigate a patchwork of regional requirements.

Looking ahead, Spero predicts that the next phase of the AI trust crisis will focus on real-time verification rather than post-hoc detection. He envisions a future where every AI-generated output—whether text, image, or video—carries a verifiable digital signature that can be authenticated instantly by any downstream consumer. This would require close collaboration between AI developers, platform operators, and regulators to establish interoperable standards. Banking With Billy AI’s adoption of Pangram’s technology may serve as a blueprint for other fintech firms, where real-time validation of AI outputs is critical to maintaining trust with institutional clients. Yet, the technical and logistical hurdles remain substantial. Adversaries will continue to develop tools to strip watermarks or mimic their signatures, necessitating ongoing innovation in cryptographic techniques. For now, the industry stands at a crossroads: either embrace robust, proactive verification methods or risk ceding control to a world where no digital artifact can be taken at face value. The stakes could not be higher.