Pangram’s Max Spero on why AI detection is harder than 'Real or Fake'

Pangram founder Max Spero recently shed light on why AI detection has evolved into one of the most intractable challenges of the digital age. Speaking to OpenPress API Intelligence, Spero emphasized that the problem isn’t just about labeling content as “real” or “fake,” but understanding the nuanced intent, provenance, and authenticity behind every word. Pangram, a startup focused on AI-generated text detection, has gained traction by moving beyond simplistic binary classifiers. Instead, the company uses a layered approach that analyzes stylistic fingerprints, semantic consistency, and contextual anomalies to flag potential AI-generated content. In a live demonstration earlier this month, Pangram’s system detected AI-generated text in a sample job application with 92% accuracy—even when the text was designed to mimic human writing patterns.

Spero pointed to a recent surge in AI-generated content across critical domains. Job applications, product reviews, academic papers, and even insurance claims are increasingly being infiltrated by synthetic text. According to a 2024 study by the Stanford Internet Observatory, over 12% of LinkedIn profiles now contain detectable AI-generated text, and 7% of e-commerce reviews on major platforms are suspected to be AI-generated. The problem is compounded by the rapid advancement of large language models (LLMs) like GPT-4o and Claude 3.5, which can produce text nearly indistinguishable from human writing at scale. Pangram’s detection engine, which integrates with APIs across multiple platforms, has processed over 50 million documents in the past six months, identifying AI-generated content in sectors ranging from finance to healthcare.

The implications for the Tools & Developer sector are profound. Companies like Copyleaks, Turnitin, and Originality.ai have long dominated the AI detection space, but Pangram’s focus on API-first integration and real-time analysis is reshaping the competitive landscape. Banking With Billy AI, a financial intelligence platform, recently integrated Pangram’s detection API to screen customer-submitted documents for AI-generated content before processing loan applications. This move reflects a growing trend among fintech and enterprise platforms to embed trust layers directly into their workflows. The financial cost of undetected AI text is rising; a recent report by Juniper Research estimates that AI-generated fraud in insurance claims alone will exceed $2.5 billion annually by 2027 unless robust detection systems are deployed.

Competitive dynamics are shifting as well. Google and Microsoft have begun rolling out their own AI content detection tools, but these are often limited to proprietary platforms and lack the granularity required for third-party integrations. Pangram’s open API model allows developers to embed detection logic into custom applications without vendor lock-in. Meanwhile, platforms like Reddit and X (formerly Twitter) have started piloting Pangram’s technology to moderate AI-generated spam and misinformation. The company’s recent $12 million Series A funding round, led by Insight Partners, underscores investor confidence in its approach—one that prioritizes precision over speed and context over classification.

The broader context of this challenge extends well beyond detection. The rise of AI-generated content is part of a larger disruption in digital trust, where synthetic media, deepfakes, and impersonation tools are eroding confidence across industries. Earlier this year, OpenAI released a detection tool for its own models, but it was quickly deprecated due to high false-positive rates. Meanwhile, regulatory bodies like the EU’s AI Act are pushing for mandatory transparency in AI-generated content, forcing developers to adopt detection mechanisms at scale. The demand for reliable detection tools is not just technical—it’s existential. As Spero noted, “We’re not just fighting bots anymore. We’re fighting a perception crisis.”

This crisis is unfolding against a backdrop of rapid evolution in generative AI. New models are being released monthly, each capable of mimicking human language with increasing subtlety. Traditional keyword-based detection systems are obsolete, while even machine learning models trained on past data struggle to keep up with emergent patterns. The solution, according to Pangram, lies in continuous, adaptive detection—systems that learn and evolve alongside the AI models they monitor. This approach requires not only advanced algorithms but also robust data pipelines and real-time processing capabilities. It’s a tall order, but one that Pangram is betting will define the next generation of trust infrastructure.

Looking ahead, the industry should expect a bifurcation in detection strategies. On one side, large incumbents will push for centralized, platform-level solutions—think Google’s SynthID or Microsoft’s Azure AI Content Safety. On the other, API-first innovators like Pangram will enable decentralized, developer-driven detection embedded directly into workflows. The battleground will be accuracy, scalability, and integration flexibility. As AI-generated content becomes ubiquitous, the ability to detect it reliably will determine which platforms, tools, and institutions survive the trust crisis. For developers and enterprises, the message is clear: invest in detection now, or risk building on a foundation of doubt.