OpenAI Tries to Shift Responsibility to Users (buttondown.com)

OpenAI quietly revised its usage policies to warn users not to rely on ChatGPT for “tailored advice that requires a license, such as legal or medical advice,” prompting debate after users feared a ban on asking such questions. OpenAI’s safety lead clarified the change wasn’t new and that model behavior is unchanged, but critics argue the company is trying to shift blame onto users for harmful or incorrect guidance. Researchers and commentators note these systems generate authoritative-sounding language that can mislead people—especially when startup hype promises to replace professionals—while the company continues to commercialize access. For the AI/ML community this highlights a core technical and ethical issue: large language models are optimized to produce fluent, persuasive text, not to validate facts or apply domain-specific standards of care or law. Studies (e.g., UBC research) show users can find LLM interactions more convincing than human professionals, raising real-world safety and liability risks. The practical implications are clear: high-stakes deployments need rigorous domain evaluation, human-in-the-loop workflows, explicit uncertainty signaling, provenance/grounding for claims, and clearer product-level accountability or regulation rather than shifting responsibility solely to end users.