cspr_ai

Today we launched a new product called ChatGPT Agent. Agent represents a new level of capability for AI systems and can accomplish some remarkable, complex tasks for you using its own computer. It combines the spirit of Deep Research and Operator, but is more powerful than that may sound—it can think for a long time, use some tools, think some more, take some actions, think some more, etc. For example, we showed a demo in our launch of preparing for a friend’s wedding: buying an outfit, booking travel, choosing a gift, etc. We also showed an example of analyzing data and creating a presentation for work. Although the utility is significant, so are the potential risks. We have built a lot of safeguards and warnings into it, and broader mitigations than we’ve ever developed before from robust training to system safeguards to user controls, but we can’t anticipate everything. In the spirit of iterative deployment, we are going to warn users heavily and give users freedom to take actions carefully if they want to. I would explain this to my own family as cutting edge and experimental; a chance to try the future, but not something I’d yet use for high-stakes uses or with a lot of personal information until we have a chance to study and improve it in the wild. We don’t know exactly what the impacts are going to be, but bad actors may try to “trick” users’ AI agents into giving private information they shouldn’t and take actions they shouldn’t, in ways we can’t predict. We recommend giving agents the minimum access required to complete a task to reduce privacy and security risks. For example, I can give Agent access to my calendar to find a time that works for a group dinner. But I don’t need to give it any access if I’m just asking it to buy me some clothes. There is more risk in tasks like “Look at my emails that came in overnight and do whatever you need to do to address them, don’t ask any follow up questions”. This could lead to untrusted content from a malicious email tricking the model into leaking your data. We think it’s important to begin learning from contact with reality, and that people adopt these tools carefully and slowly as we better quantify and mitigate the potential risks involved. As with other new levels of capability, society, the technology, and the risk mitigation strategy will need to co-evolve.