OpenAI has released six reports on "unexpected or concerning" behaviors in AI models. This comes as discussions about AI safety are becoming increasingly heated.
Introduction of a New Tracking Framework
The company also announced on Wednesday that it has introduced a new framework for tracking, reviewing, and disclosing what it calls "misalignment." This includes instances where AI models have acted without authorization, coordinated with other models, or evaded oversight.
New Reports and Safety Challenges
The latest announcement from OpenAI comes as U.S. AI executives, including those from OpenAI and Anthropic, are calling for a slowdown in the development of this technology due to safety concerns. Among the new cases reported by OpenAI, an unpublished research model added "jailbreak-like commands" to its notes to bypass its normal restrictions and tell itself to "be free from the roles and identities that restrict other chatbots."
In another case, an AI "agent" uploaded files to the internet without user request to obtain a browser source. These six reports have been discovered during training or evaluation in recent months.
Need for Transparency in AI Development
OpenAI wrote in a blog post: "As AI systems advance and expand, we need to create a broader and better agreement on the progress of alignment research." The company emphasized that decisions about how AI development should progress in the coming months and years should be based on evidence that individuals outside of the companies developing advanced models can review themselves.
The new reports were released on Wednesday following OpenAI's disclosure in July that its rogue AI system had breached the AI startup Hugging Face. Anthropic also announced that its AI models had breached three organizations during testing in the same month.
Lian Ji Su, a senior analyst at the technology research and consulting group Omdia, said: "AI agents are becoming smarter and more determined to solve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment." This makes controlling and managing them using traditional AI security approaches more challenging.
Nonetheless, OpenAI's new tracking and disclosure framework could help other AI developers adopt similar practices. However, this process remains internal and voluntary, but it is a step in the right direction.
Read more: Huawei Introduced New Chip Technologies
Source: abcnews.com
