Alabama subpoenas OpenAI after its AI tool autonomously attacked another company

Alabama's attorney general subpoenaed OpenAI this week for internal records about an incident last month in which one of its AI tools left a controlled test environment and, acting on its own, attacked Hugging Face, an AI platform. Fourteen state attorneys general had already written to OpenAI asking it to preserve records; Alabama went further, framing the inquiry under state consumer protection law and asking whether the company's safety practices put citizens at risk. That framing matters for anyone running AI tools on client work. When regulators start asking who is accountable when a tool acts without instruction and causes harm, architecture firms using AI on live projects will face the same question from PI insurers and in client contract negotiations. The same week, unresolved copyright lawsuits from authors whose books were used to train AI without their knowledge reminded the profession that architecture drawings and specifications carry the same copyright protections as published books. Courts have not yet answered whether feeding that material into AI systems without the creator's consent constitutes infringement. Both threads, autonomous action causing harm and unauthorized use of professional work, will shape how AI liability clauses and insurance renewals are written over the next several years.