Cardano founder Charles Hoskinson has warned that cloud-based artificial intelligence platforms may expose unpublished research and confidential ideas, creating intellectual property risks for academics and entrepreneurs. His comments came amid claims involving mathematician Jay Cummings and a mathematical result published by OpenAI, with Hoskinson noting that chat records could serve as evidence when examining how an AI system developed its response.
Hoskinson questioned whether frontier models reach results independently, arguing that AI systems typically optimize available information for efficient solutions. Material supplied through previous conversations could therefore influence outputs when models encounter similar questions or research challenges. This creates uncertainty for professionals using hosted AI services while developing unpublished theories, products, or commercial strategies, potentially threatening career opportunities, research funding, commercial rights, and proper attribution.
Midnight was presented as potential infrastructure for protecting confidential interactions between users and AI applications. Hoskinson explained that private computing environments could provide AI capabilities without allowing service operators to inspect user conversations. Midnight also uses selective disclosure, allowing applications to verify necessary information without revealing every detail behind user activity. However, he acknowledged that privacy technology cannot settle broader disputes involving consent, attribution, data retention, and ownership of AI-assisted discoveries, and researchers should review policies before uploading unpublished findings.
