Since my work involves AI and risk, people have been asking for my take on the OpenAI models breaking containment.

Two words: risk controls.

In more detail:

We'll probably learn more about the story in the coming days. In the meantime, you can treat this (and every other public-facing incident) as a reason to review your company's genAI practices.

• What risks have you identified?
• How often do you check?
• What controls have you established, and when did you last test them?

Every genAI model has a strain of what I call "The Random™," that wild animal that gets out and causes trouble. There's no way to 100% control it and prevent every incident, but solid risk management practices will go a long way to protecting your business, your customers, and anyone else impacted by your work.