When OpenAI released its latest usage report, it highlighted ChatGPT's applications in education and business productivity. When Anthropic published Claude's adoption metrics, the company emphasized safety-conscious deployment patterns. But researchers like Anka Reuel, a Computer Science PhD candidate, point to a fundamental problem: these narratives rest entirely on data selected and presented by the companies themselves. "There is no independent source to corroborate it," Reuel notes. This selective disclosure creates an asymmetry at the heart of AI governance. Companies decide what metrics to publish, which use cases to highlight, and crucially, which concerning patterns to omit. Without independent verification, claims about AI systems' beneficial deployment or responsible usage patterns remain largely unvalidated assertions rather than audited facts.
The regulatory landscape has not yet developed robust mechanisms to address this accountability gap. While the EU's AI Act establishes compliance requirements and NIST has published AI risk management frameworks, neither mandates independent third-party auditing of how AI products are actually used in practice. Academic researchers have largely operated in a norm-setting vacuum, lacking institutional authority or industry access to conduct systematic audits. Compare this to adjacent industries: financial firms face SEC-mandated third-party audits, pharmaceutical companies submit to FDA inspection protocols, and investment advisors undergo FINRA examinations. These sectors learned decades ago that self-reporting creates perverse incentives. Yet the AI industry has largely resisted similar mechanisms. While some companies have experimented with red-teaming and vulnerability disclosures, these remain voluntarily scoped and controlled. Formal auditing requests have been rare, and when researchers have sought deeper access to training data or deployment analytics, companies have typically declined or severely restricted scope.
The stakes extend beyond corporate transparency into public understanding itself. When most people learn about AI capabilities and usage patterns, they rely on company-curated narratives that may systematically understate risks or overstate benefits. This information asymmetry shapes policy discussions, investor expectations, and public sentiment. As AI systems increasingly influence consequential decisions—from hiring to law enforcement to healthcare—the absence of independent auditing creates a credibility problem that ultimately threatens industry legitimacy. Without third-party mechanisms comparable to those in finance or pharmaceuticals, policymakers lack reliable ground truth about how these systems function at scale. The path forward requires not merely company transparency initiatives but mandatory, independently-conducted auditing standards with genuine enforcement authority and public reporting requirements.