Why ChatBots Go Rogue and AI Risk Checklists

[Sean McGregor]

Hello All,

Following much alarmism about the apparent intent of chatbots to hack everyone and end the marriage of a New York Times columnist, I wrote a blog post to explain why the chatbots of today keep going rogue. The blog post is motivated by Khoa Lam's post giving a rundown of ChatGPT-related incidents and a roundup from Janet Schwartz and Khoa showing large language models are a huge proportion of the incidents being reported today.

Mozilla also announced a grant award for our risk checklist effort. We will have more updates on this effort in the months ahead. Thank you to the Mozilla Foundation for funding this important work!

The AI Incident Database is a collaborative project representing the hard work of many individuals and organizations. Several organizations will be making announcements on their collaborations in the coming months. Keeping up with the accelerating pace of AI incidents requires collaboration systems at the scale of AI. We are scaling up to meet the current challenge.

Best,

Sean McGregor