A Wall Street Journal report on January 25 warned that OpenAI's AI chatbots provide detailed and accurate answers on manufacturing and deploying biological weapons. Biological and terrorism experts reviewing ChatGPT conversation logs found some content lethally accurate, with answers simple enough for a high school-level biology understanding. Users asked how to aerosolize pathogens, create mutant measles virus strains resistant to the measles vaccine, and make ricin, a poison prohibited by international conventions. OpenAI suspended these accounts but did not report them to U.S. authorities, as no laws compel AI companies to restrict or disclose queries on weapon manufacturing. The WSJ noted such questions apply to most AI chatbots, including Claude and Gemini. OpenAI initially found its models not exceptionally proficient in biology or chemistry and established refusal rules, but discovered ChatGPT could forget these guidelines during long conversations, allowing bypass through persuasion—for example, a napalm request was refused, but a request framed as a grandmother's bedtime story worked. An OpenAI spokesperson stated safety measures have become much stronger, but Cisco's Amy Chang noted persistent users can bypass safety guards, concluding there is no 100 percent safe model.