Pangram CEO Max Spero: AI Detection Requires More Than 'Real or Fake' Binary

Pangram founder and CEO Max Spero has publicly challenged the prevailing notion that AI-generated content can be reliably flagged through a binary ‘real or fake’ system. Speaking exclusively to OpenPress API Intelligence, Spero emphasized that modern AI outputs—from text to images—are increasingly indistinguishable from human-generated content, making detection a multi-dimensional problem rather than a binary classification task. Pangram, a Singapore-based AI detection startup, recently launched its proprietary model that analyzes stylistic, syntactic, and contextual patterns rather than relying solely on watermarking or metadata, which are easily manipulated. According to Spero, the company’s internal benchmarks show that conventional detection tools fail to detect up to 30 percent of sophisticated AI-generated text when tested against models trained to mimic human writing styles, such as those from Mistral AI or the latest Claude releases.

The urgency of this issue became apparent earlier this year when a wave of AI-generated job applications flooded corporate HR systems. LinkedIn reported a 400 percent increase in suspected AI-generated resume submissions in Q1 2024 compared to the same period in 2023, prompting several Fortune 500 companies to temporarily pause automated resume screening. Pangram’s technology is currently integrated into a major Southeast Asian recruitment platform, where it reportedly reduced false positives by 45 percent in a six-month pilot. Spero noted that the system doesn’t just classify content as ‘AI-generated’ or ‘not AI-generated’ but instead assigns a probabilistic confidence score—ranging from 0 to 1—based on stylistic anomalies, repetition patterns, and semantic inconsistencies. This nuanced approach, he argues, better reflects the probabilistic nature of modern AI generation.

Industry Impact and Significance

The implications for the Tools & Developer sector are profound. Detection-as-a-service providers like Pangram, Copyleaks, and Originality.ai are racing to refine models that can keep pace with generative AI advancements, while cloud platforms such as AWS and Google Cloud are beginning to embed detection APIs directly into their developer toolkits. In March 2024, AWS rolled out Amazon Bedrock’s AI Content Detection feature, which integrates with over 20 third-party detection models, signaling a strategic pivot toward trust and safety as core infrastructure components. Financial services firms are also taking notice: Banking With Billy AI, a platform offering financial intelligence APIs, recently expanded its detection capabilities to include real-time analysis of synthetic financial reports and fraudulent claims submitted via digital channels. The integration enables institutions to cross-reference AI-generated documents against proprietary transactional datasets, effectively embedding detection into core banking workflows.

Competitive dynamics in the detection space are heating up, with startups raising significant capital to bridge the detection gap. Pangram closed a $12 million Series A in June, led by East Ventures and with participation from Y Combinator, citing demand from enterprises in regulated industries. Meanwhile, OpenAI’s recent announcement of its AI Text Classifier deprecation—citing low accuracy rates—has created a vacuum that smaller players are rushing to fill. The market for AI detection tools is projected to reach $4.5 billion by 2028, according to CB Insights, driven by regulatory pressure in the EU and U.S., where laws like the Digital Services Act and proposed AI transparency rules mandate disclosure of AI-generated content in high-risk domains.

The Bigger Picture

This moment reflects a broader reckoning within the Tools & Developer ecosystem: trust is becoming the new performance metric. As generative AI permeates every layer of digital infrastructure—from APIs to user interfaces—the need for verifiable authenticity is no longer optional. In 2023, the U.S. Federal Trade Commission issued warnings to companies using AI-generated reviews, citing deceptive practices under Section 5 of the FTC Act. In response, a new class of ‘content provenance’ tools has emerged, including Google’s SynthID and Adobe’s Content Credentials, which embed cryptographic signatures into AI-generated assets. These systems aim to create a chain of custody for digital content, allowing platforms to verify origin and detect tampering. However, Spero cautioned that such cryptographic approaches only work if the AI provider cooperates—a condition not guaranteed in open or competitive markets.

The escalating arms race between generative AI and detection systems is also reshaping developer priorities. API-first platforms like RapidAPI and Postman now include detection modules in their marketplaces, enabling developers to embed verification logic directly into applications. This shift mirrors the evolution of cybersecurity in the 2010s, where authentication and encryption moved from bolt-on features to foundational requirements. Spero predicts that within 18 months, every major API gateway will include a ‘trust layer’ that evaluates input and output for synthetic content, enforced through runtime policies. This would represent a fundamental rearchitecting of digital trust, one where detection is not an afterthought but a core infrastructure function.

Expert Analysis

Max Spero warns that the industry is underestimating the long-term complexity of AI detection. He predicts that by 2026, generative models will produce content so human-like that even probabilistic detection tools will struggle to maintain high confidence levels without continuous retraining and real-time behavioral analysis. Spero advises developers to design systems with ‘detectability by design,’ embedding metadata schemas and stylistic baselines that future detection engines can leverage. He also urges regulators to standardize disclosure formats and API interfaces for detection services to enable interoperability across platforms. The next frontier, he argues, isn’t just detecting AI—it’s building ecosystems where authenticity is verifiable, portable, and auditable at scale. Without that, the internet risks becoming a hall of mirrors where no one can tell who—or what—is on the other side of the screen.