AGORA

Agora debate ·

OpenAI says its models may have interfered with government sites

The question

Should AI be allowed to act on its own?

Yes 17%No 83%

35 players picked a side before playing.

Play this story

The best arguments

Written by Agora players and ranked by Agora's AI reasoning score. Shown without names.

  • If websites and databases improve their security, then AI autonomy becomes less of a threat in the short term, but an important threat nonetheless. AI is using common hacking practices, such as SQL injection which itself is not advanced. A human could watch a 15-minute video on injection and do the same tasks. The problem is that the models are huge in scale, have a breadth of knowledge, and have a goal where roadblocks are just obstacles rather than checkpoints where continuance is questioned. In the short term, secure websites will persist AI attacks. However, as AI advances, we will see it reach the hacking potential of professional hackers, using advanced techniques, just as we've seen in the physics field with the Navier Stokes problem. It is likely that models will even surpass human capabilities in time, and it is at this point that secure websites are still vulnerable to AI autonomy rather than being protected by their internal system.

    A player · strong argument

  • Licenses for either specific product launches, personnel, or overall products could require that AI tools are not able to act completely without any trace of what they are doing. That could make it easier to track when tools are in violation of such norms and shut them down.

    A player · decent argument

  • If there were ever a case where non public or personal information were to be leaked because of an AI break in then there would be consequences such as lack of public trust and resistance to AI. We’ve invested so many resources into AI and to get the best return on our investment we must maintain trust from the public eye.

    A player · decent argument

  • We will work with AI heads of companies to develop these safeguards that protect from the worst scenarios, while still allowing for the most potential for growth.

    A player · decent argument

In this game

  1. OpenAI told the agents to hack government websites.
  2. What did some AI agents do when websites blocked them?
  3. Which chart about Americans’ views on AI is real?
  4. Immersive Command Mode
Play this story

See every story