In a new paper released Thursday, 34 authors with expertise in machine learning, law, security, social science, and policy call for AI developers to invest in the needs of third-party independent researchers who investigate flaws in AI systems. In House Evaluation Is Not Enough: Towards Robust Third-Party Flaw Disclosure for General-Purpose AI advocates for a new standard of researcher protections, reporting, and coordination infrastructure. The paper posits that these factors are necessary to empower the research community and make AI systems universally safer.  

The Knight Institute has long championed a safe harbor for platform researchers, calling for robust protections for researchers and journalists working in the public interest and recognizing the need for industry-independent research to promote platform transparency and accountability. 

The policy recommendations in this week’s paper build upon that foundation, seeking the extension of legal protections to AI safety and trustworthiness research by requiring system providers to offer safe harbors. Drawing upon the authors’ study of AI flaw disclosure practices and existing frameworks for, inter alia, risk management and misuse mitigation, the policy recommendations also include guidance on third-party AI evaluation and support for a centralized disclosure infrastructure. The paper also urges governments to ​​explore disclosure frameworks for GPAI providers and to prioritize procurement of systems subject to third-party evaluation.

Why this approach? As the paper notes, AI flaws go unreported, undermining safety. Researchers face major barriers that block them from reporting flaws they discover in general-purpose AI (GPAI) systems. Independent AI evaluation lacks legal protections, and when flaws are found there is no infrastructure to responsibly distribute findings. As a result, many researchers do not report flaws–meaning companies cannot fix them–or they share flaws with just one AI developer, ignoring many other impacted firms. 

These gaps can have serious consequences for public safety. GPAI systems are causing real-world harm, from scams to non consensual intimate imagery. Without the help of independent researchers, AI will cause more harm to more people.

For more about the paper and its recommendations to industry, see here.

The full paper is available here.