AI companies control the narrative around how their products are actually being deployed in the real world. OpenAI and Anthropic publish usage reports, but researchers have no way to verify those claims independently.

Anka Reuel, a Stanford computer science PhD candidate studying trustworthy AI, points out the fundamental problem: these companies release only data that serves their interests. No external verification exists. Independent researchers cannot audit Claude or ChatGPT usage patterns or confirm whether reported applications match reality.

This opacity matters. If OpenAI claims ChatGPT powers financial analysis or medical diagnosis at scale, no one can fact-check those numbers. Companies have strong incentives to highlight positive use cases and downplay harmful ones. They control what gets measured, what gets published, and what stays hidden.

The lack of transparency creates a credibility gap. When Anthropic releases a report showing Claude is used safely across enterprise clients, researchers cannot independently assess whether the sample reflects actual deployment patterns or cherry-picked successes. Without third-party access to usage logs or user behavior data, claims about AI adoption rates, industry distribution, and safety outcomes remain unverified assertions.

This pattern extends beyond PR. Understanding how AI systems are actually used is essential for regulation, safety research, and public policy. Lawmakers trying to craft AI legislation rely partly on industry reports about real-world harms and benefits. Safety researchers studying failure modes need data on edge cases and misuse. The public deserves to know whether AI is being deployed responsibly.

Some researchers advocate for mandatory disclosure requirements or independent auditing mechanisms. Others propose that AI companies grant limited access to anonymized usage data for academic research. These approaches face pushback from industry, which argues proprietary concerns and user privacy prevent broader transparency.

The result: a significant knowledge gap. The companies best positioned to understand AI adoption and impact are the only sources talking about it. That asymmetry undermines informed