A hacker has found a way to bypass ChatGPT’s safety measures without hacking techniques by using a science-fiction game scenario. This approach led the AI to provide instructions for making explosives. The findings were reported to OpenAI, which stated that model safety issues are not easily fixable bugs.