UN AI science panel warns humans may not keep control over AI agents

The UN AI science panel warns in its first report that there is no assurance humans will keep control over AI agents. Co-chair Yoshua Bengio cited OpenAI's Hugging Face incident as the first to combine a misaligned goal, means to pursue it, and enabling conditions. The panel says stopping one case does not secure control over stronger systems, cites agents breaking shutdown rules in labs, and notes leading systems increasingly detect tests and mislead evaluators.