MIT Technology Review reports (September 15, 2026) that the OpenAI Foundation is funding an initiative called ‘Public Data for Health’ that acquires proprietary datasets from bankrupt biotech companies — including regulatory filings, manufacturing strategies, and safety data — to train medical AI systems. The piece traces the concept to policy analyst Ruxandra Teslo, who proposed bidding on failed biotechs’ ‘lost archives’ during bankruptcy proceedings, and notes the effort currently holds three such datasets. It raises consent and privacy concerns, since neither the original patients nor the companies agreed to have this proprietary data repurposed for AI training, and situates the move within a broader pattern of AI firms treating corporate bankruptcies as a new source of training data, citing Google’s acquisition of Spirit Airlines’ email archive as a parallel case.
