Microsoft study showed the risks of delegating document work to LLMs

Microsoft released clarifications to its study on how language models distort documents when tasks are delegated to them. The article explains what the team actually demonstrated — and where critics misinterpreted the results. The work raised an important question about the reliability of AI in long-running processes.