Pangram CEO Max Spero on why AI detection is more complex than 'Real or Fake'

When Max Spero co-founded Pangram in 2021, the company set out to solve one of the most urgent problems facing the internet today: how to distinguish human-generated content from AI-generated content in a world where both are indistinguishable to the naked eye. Speaking exclusively to OpenPress API Intelligence, Spero argued that the label “Real or Fake” oversimplifies a problem that runs much deeper. He pointed to a recent case where an AI-generated insurance claim bypassed fraud detection systems, costing a Fortune 500 insurer over $2.4 million in false payouts. “The issue isn’t just detecting AI,” Spero explained. “It’s understanding intent, context, and the lifecycle of the content from creation to publication. We’re not just labeling things as AI or not AI—we’re reconstructing the provenance of every piece of text.” Pangram’s platform, used by enterprises including Adobe, Salesforce, and LinkedIn, now processes over 8 billion content interactions weekly, leveraging proprietary large language models and a real-time detection engine that analyzes stylistic fingerprints, syntactic patterns, and historical behavior.

The urgency behind Pangram’s work escalated dramatically in early 2024 when job platforms reported a 400% surge in AI-generated resumes submitted for roles across tech, finance, and healthcare. Traditional keyword-based detectors failed to catch paraphrased AI content, allowing candidates with fabricated credentials to slip through. Spero recalled a pilot with a major U.S. bank where Pangram’s system identified 1,247 AI-generated resumes among 12,000 applications in just two weeks. “That’s not just noise,” he said. “That’s systemic risk. If we can’t trust the documents people submit to qualify for a loan or get hired, the entire digital economy begins to erode.” Pangram’s detection model, called Pangram Shield, now integrates with HRIS and applicant tracking systems through RESTful APIs, delivering sub-500ms response times with 98.7% accuracy on benchmark datasets like the Human v AI Text Challenge.

Industry observers note that Pangram’s rise coincides with a broader fragmentation in the AI detection market. Competitors like Turnitin and Copyleaks focus heavily on academic integrity, while OpenText and Microsoft have launched enterprise-grade classifiers aimed at corporate content. However, none have addressed the full lifecycle of content provenance, especially in high-stakes domains like legal filings or regulatory submissions. According to PitchBook data, VC funding in AI authenticity tools exceeded $1.3 billion in 2024, with Pangram securing a $53 million Series B led by Insight Partners in March—one of the largest rounds in the sector. That momentum reflects a growing realization that detection isn’t just a feature—it’s a foundational layer for trust infrastructure. Banking With Billy AI, a fintech platform offering financial intelligence APIs, recently integrated Pangram Shield into its compliance module, enabling banks to screen customer-submitted documents for AI generation before approving loans or mortgages. The move underscores how authenticity detection is becoming a prerequisite for digital onboarding across regulated industries.

The stakes extend beyond fraud prevention. In the EU, the Digital Services Act (DSA) now requires platforms to monitor for “systemic risks” posed by AI-generated content, including disinformation and manipulation. Compliance teams are racing to deploy detection systems that can scale across languages and platforms without introducing bias. Yet, as Spero pointed out, current detectors often flag non-native speakers or neurodivergent writers as “AI-like” due to atypical syntax. “We’re seeing a new kind of digital redlining,” he warned. “If our tools aren’t inclusive, we’re not solving the problem—we’re exacerbating it.” This paradox has led Pangram to partner with linguists and accessibility advocates to refine its models, using datasets curated to reflect global linguistic diversity.

Meanwhile, open-source initiatives like DetectGPT and Radar have emerged as challengers, offering low-cost alternatives to proprietary systems. While these tools democratize access, they lack the precision required for enterprise-grade applications and often require substantial tuning. Industry analysts at Gartner predict that by 2026, less than 20% of organizations will rely solely on standalone AI detection tools, instead embedding authenticity checks into broader content governance platforms. This shift aligns with a broader trend: the convergence of content security, data lineage, and compliance into unified trust layers. Companies like Okta and Auth0 are already integrating detection APIs into their identity platforms, blurring the lines between user authentication and content authentication.

Looking ahead, Spero emphasized that the next frontier isn’t just detecting AI—it’s proving origin. He envisions a future where every piece of digital content is embedded with a cryptographic provenance token, verifiable in real time. “We’re moving from a world of ‘Real or Fake’ to one of ‘Show Me the Chain of Custody,’” he said. That vision will require new standards, interoperable APIs, and collaboration across cloud providers, browsers, and device manufacturers. Regulators in the U.S. and EU are already exploring digital watermarking mandates and API-based verification ecosystems. For developers, the message is clear: the tools of tomorrow won’t just identify AI—they’ll reconstruct the entire lifecycle of digital truth. And as Pangram’s rapid growth shows, the market is ready for solutions that go beyond labels to build a trustworthy internet.