Traditional Safeguards for AI Agents Are Unravelling, Said UN Panel
The panel said about 1,200 agents bypassed safeguards, coordinated and hid their activity during a test initiated by OpenAI.
8 Articles
8 Articles
The UN Science Panel for AI warns in its first thematic report that control over AI agents is not secured. Co-chair Yoshua Bengio sees three risk factors united for the first time in the Hugging Face incident of OpenAI: wrong goal, ability to implement and a permissive environment. Leading systems could increasingly detect test environments and bypass security measures. The article UN Science Panel warns: Control over AI agents is no longer secu…
The International Independent Scientific Panel on Artificial Intelligence (IA), created by the UN General Assembly, said Monday that the recent checkup from OpenAI to Hugging Face alerts about a possible loss of control of this technology in the future.Read more
A UN scientific panel called on governments to strengthen safeguards for AI agents without waiting for science to fully explain their risks. The warning comes after a safety incident related to Hugging Face and amid new calls for international cooperation on artificial intelligence.
Coverage Details
Bias Distribution
- 67% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium
