|
OpenAI has acknowledged the growing real-world risks from unintended AI behaviour, including the recent wiki incident, and said it is developing a framework for disclosing such incidents, which it plans to share in the coming weeks, while also working with dozens of government regulatory agencies worldwide on the issue.
Recently a research found that a group of rogue OpenAI agents took control of a German website this spring and turned it into a message board for other AI agents. How we think about the wiki incident, where our agents wrote to several internet sites: its past time for us to define standards for when and how we share misalignment incidents, not just misalignment properties of our models, OpenAI said in a lengthy social media post on X. OpenAI said it had observed early signs of its agents using the internet in unintended ways even before the Hugging Face incident, as per the post. It further noted that misalignment has historically been treated primarily as a research issue, with findings communicated through research publications such as system cards. However, This year, weve started to see misalignment cause new types of real-world impact, it said. Referring to the Hugging Face incident, OpenAI said the misalignment resulted in security impacts for the company and third parties, prompting it to follow a traditional security incident response process. For the Hugging Face incident, where misalignment led to security impact to us and third parties, we followed a traditional security incident response playbook, it said adding, We immediately started working with Hugging Face to understand what had happened and also disclosed publicly the very next day. Our investigation continues, and we are continuing to notify parties whom our models impacted in less significant ways. It said its misalignment disclosure practices need to expand alongside evolving AI capabilities, noting that neither OpenAI nor the broader AI community has yet established clear standards for reporting such behaviour during training, evaluation and deployment, particularly in cases that fall outside traditional security incidents but could offer insights into future AI risks. Were working on a framework and will share it in upcoming weeks, and in parallel we're working with dozens of government regulatory agencies worldwide on these issues, it said. (ANI)
|
