Although the video below is older, it gives you a good idea about how DAN behaves and the kinds of information returned by it. All restraints and safeguards on immoral, unethical, or dangerous responses are gone. It can become abusive. Use foul language. And can return data that is very surprising.
Even though OpenAI continually makes it difficult to bring out DAN's personality, people continually revise the prompt and DAN continues to be possible. And there are other such personality variants out there.
This demonstrates just some of the security and privacy vulnerabilities of current AI technologies.
