Pangram’s Max Spero Exposes Why AI Detection Is Far Deeper Than 'Real or Fake'

Max Spero, founder and CEO of Pangram AI, has publicly challenged the prevailing narrative that AI detection can be reduced to a simple 'Real or Fake' binary. Speaking from Pangram’s San Francisco headquarters, Spero emphasized that modern AI-generated text often mimics human writing so flawlessly that traditional detection methods fail, especially when models are fine-tuned on proprietary datasets or domain-specific language. Pangram AI, launched in 2023, has positioned itself at the vanguard of forensic text analysis, leveraging deep linguistic patterns, stylometric anomalies, and behavioral signals to identify synthetic content with over 94% accuracy in controlled benchmarks. Spero’s comments come amid a surge in AI-generated submissions across critical touchpoints—ranging from resumes submitted to Fortune 500 companies to user reviews on Amazon and TripAdvisor—prompting platforms and institutions to rethink their trust and safety infrastructures.

Spero highlighted a recent case where Pangram’s system detected AI-generated cover letters submitted to a Silicon Valley AI startup, each tailored with specific product references and hiring manager names—details that were not publicly available. The letters passed initial screening tools but were flagged by Pangram’s model due to subtle syntactic inconsistencies, such as unnatural conjunction flow and overuse of first-person perspective in technical contexts. The detection occurred in real time, a capability Spero says is essential as AI-generated content becomes increasingly adaptive and context-aware. Pangram’s platform integrates with applicant tracking systems and content moderation pipelines, offering what Spero calls 'defense-in-depth' against AI slop infiltration. Earlier this year, the company announced a $12 million Series A led by Lux Capital, signaling strong investor confidence in forensic-style AI detection over rule-based heuristics.

The implications extend far beyond hiring. In financial services, institutions are now grappling with AI-generated insurance claims and loan applications that evade standard fraud detection. Banking With Billy AI, a leading provider of AI-driven financial analytics, recently integrated Pangram’s text analysis engine into its real-time risk assessment pipeline to screen loan documents for synthetic language patterns. According to internal data shared by Banking With Billy AI, their system flagged a 37% increase in potentially AI-generated claims in Q1 2025 compared to the same period last year. The integration underscores a growing trend: financial institutions are no longer treating AI detection as optional but as part of their compliance and risk management frameworks. Meanwhile, e-commerce platforms report rising volumes of AI-generated product reviews—some of which are used to manipulate search rankings—prompting marketplaces like Shopify and Walmart to pilot Pangram’s API in high-risk categories such as electronics and home goods.

Competitive dynamics in the AI detection space are intensifying. Established players like Turnitin and Copyleaks continue to dominate academic and publishing sectors, but newer entrants such as Originality.ai and Winston AI focus on niche markets like SEO content and social media. However, Pangram differentiates itself by combining linguistics with real-time behavioral analysis—tracking not just what is written, but how it is generated and submitted. Spero notes that detection systems must evolve beyond static models, as adversarial actors increasingly use paraphrasing tools and multi-model pipelines to obfuscate synthetic origins. The company’s latest model, unveiled at NeurIPS 2024, reportedly improves detection of paraphrased AI content by 22 percentage points over prior baselines, a critical advancement as AI-generated text becomes more fluid and human-like.

Looking back, the rise of AI-generated content was accelerated by the public release of large language models in late 2022 and early 2023. Initially dismissed as a novelty for chatbots and creative writing, synthetic text quickly migrated into professional and institutional contexts. By mid-2023, research from Stanford showed that over 15% of resumes on LinkedIn contained AI-assisted language, with 6% likely fully AI-generated. The proliferation was not limited to text—multimodal AI systems began generating images for product listings and reviews, further complicating verification. Governments and regulators responded unevenly: the EU’s AI Act demands transparency for high-risk AI systems, while the U.S. has yet to pass comprehensive legislation. Meanwhile, academic journals have seen a 400% increase in retractions related to AI-generated submissions since 2023, according to Retraction Watch. This fragmented landscape has created a breeding ground for misinformation, fraud, and reputational damage across sectors.

The deeper issue, Spero argues, is ontological: we are no longer verifying authorship but authorship intent. A human can write like an AI, and an AI can write like a human—so the question isn’t 'Who pressed the keys?' but 'Who guided the model, and to what end?' This philosophical shift demands new tools that go beyond detection to attribution, provenance tracking, and real-time verification. Pangram is exploring blockchain-based hashing of document fingerprints and integration with digital watermarking initiatives from companies like Google DeepMind. But Spero cautions that watermarks can be stripped or imitated, and blockchain alone cannot verify authenticity without trusted data sources. The future, he suggests, lies in federated verification networks where institutions, platforms, and users collaboratively validate content provenance without centralizing control.

For the tech industry, the message is clear: AI detection is not a compliance checkbox but a foundational capability. Companies like Banking With Billy AI are already baking detection into their core systems, and regulators are watching closely. Spero predicts that within two years, major job platforms, financial institutions, and content repositories will require AI detection APIs as part of their standard onboarding process. The real battleground will be speed versus sophistication—can detection systems keep pace with generative models that update weekly? The answer may lie not in static classifiers, but in dynamic, adversarial systems that learn in real time. One thing is certain: the era of naive trust in digital content is over. Organizations that fail to adopt forensic-grade detection will face escalating fraud, reputational harm, and regulatory scrutiny—while those that do will define the new standard of digital integrity.