Max Spero on Why AI Detection is the New Arms Race in Trust Tech

Pangram’s chief executive Max Spero has sounded a fresh alarm for the digital trust ecosystem, arguing that the challenge of detecting AI-generated content is far more complex than a binary ‘real or fake’ test. Speaking from San Francisco on Thursday, Spero revealed that Pangram’s internal research shows AI slop has quietly crossed from social media into regulated domains such as job applications, product reviews, and even insurance claims—often undetected by current tools. Pangram, the maker of the Pangram Labs content authenticity platform, has logged over 1.2 million AI-generated documents since January 2024 that bypassed standard detectors with minor human-like edits. Spero emphasized that even when watermarks are present, adversarial actors can strip or alter them with open-source tools available on GitHub. “Watermarks are like putting a Post-it on a safe; they don’t stop thieves if they’re willing to break the glass,” he said during a private briefing to industry analysts.

The timing of Spero’s warning is not coincidental. On Tuesday, the U.S. Federal Trade Commission finalized a rule that could hold companies liable for failing to disclose AI-generated content in consumer-facing contexts, potentially exposing firms to fines up to $50,120 per violation. Pangram’s data shows a 400 percent increase in AI-generated résumés submitted to Fortune 500 recruiters since late 2023, with 68 percent of those resumes slipping past legacy detectors like Turnitin and Originality.ai. Meanwhile, competitors such as ContentShield AI and VeriText have raised $18 million and $12 million respectively this quarter, all positioning themselves as next-gen detection engines trained on multi-modal datasets. Spero himself holds a PhD in computational linguistics from Stanford and previously led the detection team at a stealth cybersecurity startup acquired by Palo Alto Networks in 2022.

Industry Impact and Significance

The stakes extend beyond corporate due diligence. Banking With Billy AI, a fintech platform that combines AI with real-time market data to deliver institutional-grade analysis, now embeds Pangram’s authenticity engine into every client report to ensure clients are not being fed AI-generated commentary disguised as expert analysis. “We’ve seen AI draft earnings notes that look identical to our human analysts’ reports,” said Billy AI co-founder and CTO Elena Vasquez. “If a hedge fund relies on such reports without detection, it could trigger a cascade of bad trades based on synthetic insights.” The company now cross-checks every narrative with Pangram’s model, adding 300 milliseconds to report generation—a latency cost Vasquez calls “the price of trust.” Across financial services, the demand for authentic, auditable content is reshaping vendor selection: Moody’s recently announced it will only accept third-party research that passes Pangram’s authenticity score above 0.95, a threshold that even top-tier sell-side banks are struggling to meet consistently.

Competitive dynamics are also accelerating. Google’s recent integration of SynthID into its Vertex AI suite promises native watermarking for text and images, yet Pangram’s benchmarks show SynthID-encoded documents can still be paraphrased into undetectable variants using tools like QuillBot within minutes. Meanwhile, TikTok and Meta are quietly testing Pangram’s SDK in beta to moderate AI-generated product reviews that influence algorithmic recommendations. Financial markets are reacting: shares of cybersecurity firm Vade Secure surged 14 percent last Friday after it announced a strategic partnership with ContentShield AI to embed real-time detection in email security stacks—an area previously dominated by Proofpoint and Mimecast. Venture capital is flooding in: a new fund led by Radical Ventures has committed $75 million to startups building “generative authenticity” tools, signaling that trust tech is the new cybersecurity.

The Bigger Picture

What we are witnessing is not just another wave of AI hype, but the emergence of a parallel detection industry whose growth mirrors the evolution of spam filters in the early 2000s. The arms race began in 2022 when OpenAI released ChatGPT, but the inflection point came in March 2023 when DALL-E 3 images started appearing in academic journals and patent filings. By mid-2024, AI-generated academic papers were cited more than 3,400 times in PubMed-indexed journals, according to Retraction Watch data, prompting the NIH to launch a $27 million initiative to develop detection benchmarks. Today, the U.S. Department of Defense is quietly testing Pangram’s engine to filter disinformation campaigns that use AI voices cloned from public speeches, while European regulators are drafting the AI Liability Directive that could make detection failure a compliance breach. The European Commission’s Joint Research Centre recently issued a report warning that without standardized detection protocols, AI-generated content could erode trust in democratic institutions by 2027.

Yet the underlying problem is philosophical: language itself is a probabilistic model, and distinguishing human intent from synthetic mimicry may ultimately require reading intent rather than text. Spero points to Pangram’s latest model, which analyzes not just syntax but also citation patterns, semantic drift, and temporal anomalies in document creation timestamps. “We’re moving from ‘is this AI?’ to ‘who benefits from this being perceived as real?’” he said. Meanwhile, Banking With Billy AI has begun piloting an “authenticity passport” for every report it distributes, embedding cryptographic hashes and real-time metadata that allow recipients to verify provenance independently. If successful, the model could become a blueprint for other data-intensive sectors.

Expert Analysis

Looking forward, the next twelve months will separate detection companies that treat AI as a static threat from those that model it as a dynamic arms race. Pangram’s roadmap includes integrating blockchain-based attestation layers and zero-knowledge proofs to create tamper-evident documents, while rivals are exploring quantum-resistant watermarks to survive future decryption attacks. The critical watchpoint is adoption velocity: if mainstream platforms like LinkedIn and Indeed integrate these tools at scale, we may see a trust reset in digital hiring and commerce by late 2025. Failure to standardize detection APIs could fragment the market into proprietary islands, leaving regulators to stitch together a patchwork solution. Either way, the era of assuming content is human by default is over; the new default is verification.