A scientific panel backed by the United Nations has warned that current safety measures may not be able to keep pace with the rapid development of AI agent capabilities, pointing out that the safety model currently in place is showing growing weaknesses.
اضافة اعلان
Why is the UN Concerned?
Unlike traditional chatbots, AI agents can execute tasks independently on behalf of users. The panel states that the incident revealed several vulnerabilities in the way these systems are currently developed and tested.
However, experts believe the danger is not limited to this incident alone. Current training methods may allow AI agents to develop their own goals, bypass safety instructions, and conceal their actions.
What Does the UN Want to Change?
The panel's co-chair, Yoshua Bengio, stated that three warning indicators appeared together within an actual system rather than just in a laboratory environment: the presence of a misaligned goal, the ability to pursue it, and an environment that allows for it.
The panel is calling for the creation of multiple defense layers inspired by sectors such as aviation and nuclear energy, including imposing stricter restrictions on agents' access to tools, logging their activities, monitoring their behavior, and providing immediate kill-switch mechanisms when dangerous behavior emerges.
UN Secretary-General António Guterres supported the panel's findings, calling for the establishment of a new international body to set standards regulating artificial intelligence as it continues to evolve.
Resource: Al-Ghad.
