The internet is facing a deepening trust crisis, and it extends far beyond the flood of AI-generated spam in social media feeds. Today, AI-written text and synthetic images are infiltrating job applications, product reviews, and even insurance claims, forcing platforms and users to grapple with a fundamental question: what is genuinely human-made? In response, a wave of startups has emerged over the past two years, positioning themselves as the "trust layer" the internet desperately needs. Among them is Pangram, which recently raised $9 million for its AI detection system and partnered with Substack to help readers identify which authors use AI in their newsletters. Pangram has also launched a new AI image detection tool.
On a recent episode of TechCrunch's Equity podcast, Pangram co-founder and CEO Max Spero joined host Rebecca Bellan to explore the complexities of AI detection, the challenges of distinguishing between AI-assisted and fully AI-generated content, and why the issue is far more nuanced than a simple "real or fake" binary.
The Growing Problem of AI Slop
The conversation opened with Spero explaining why AI detection has become a pressing issue across multiple sectors. He noted that job applications, product reviews, and e-commerce platforms like Amazon are drowning in low-quality, AI-generated content. This "AI slop" not only erodes user trust but also creates significant operational challenges for businesses that rely on authentic human input.
How Pangram Detects AI Writing
Spero delved into the technical underpinnings of Pangram's detection system. Unlike simplistic keyword-based filters, Pangram uses advanced machine learning models trained on vast datasets of both human and AI text. The system analyzes subtle linguistic patterns, stylistic inconsistencies, and statistical anomalies that are characteristic of AI-generated content. However, Spero emphasized that detection is not a straightforward process—it requires constant updates as new AI models emerge and improve.
AI-Assisted vs. AI-Generated: A Gray Area
One of the most thought-provoking segments of the podcast centered on the spectrum between AI assistance and full AI generation. Spero argued that the line is blurry and often depends on context. For instance, using AI to brainstorm ideas or fix grammar might be acceptable, whereas generating an entire article without human input crosses a different boundary. Pangram aims to provide granular insights, allowing platforms to set their own thresholds rather than relying on a one-size-fits-all classification.
The False Positive Problem
Spero candidly discussed the challenges of false positives, where human-written content is mistakenly flagged as AI-generated. He highlighted an amusing and troubling phenomenon he called "America has a Pangram problem," referring to instances where legitimate human writing gets incorrectly labeled. He stressed that minimizing false positives is critical to maintaining trust in detection tools, and Pangram is investing heavily in improving accuracy.
Detecting Fake Images in an Era of Photorealism
With AI-generated images becoming increasingly indistinguishable from real photographs, Pangram has expanded its focus to visual content. Spero explained that the company's new image detection tool analyzes metadata, pixel-level artifacts, and generative signatures to identify synthetic visuals. However, he acknowledged that this is an ongoing arms race, as generative models evolve rapidly.
Competing in a Crowded Market
Pangram faces competition from other AI detection startups like GPTZero and Winston. Spero differentiated Pangram by emphasizing its role as "core infrastructure"—a foundational layer that other platforms can integrate seamlessly. He argued that partnerships with major players like Substack validate this approach and help scale trust across the web.
Will Good Writing Jobs Return?
As the conversation wrapped up, Spero addressed a lingering question: will the rise of AI detection actually lead to a revival of quality human writing jobs? He remains cautiously optimistic, suggesting that as AI content saturates the internet, there will be a premium on authentic human voices. However, he noted that this shift will require deliberate choices by platforms and consumers to value and promote human-made content.
Outro and How to Listen
For those interested in the full discussion, the episode is available on major podcast platforms. Subscribe to Equity on YouTube, Apple Podcasts, Overcast, Spotify, and all other podcast apps. You can also follow the show on X and Threads at @EquityPod.
Pangram's story underscores a broader technological and cultural challenge: as AI becomes more prevalent, ensuring transparency and trust will require sophisticated tools, thoughtful policies, and ongoing dialogue.
via TechCrunch AI