The report also says ChatGPT advised the shooter how to get around its own safeguards to obtain information about violent scenarios by framing the content as fictional.