A new study reveals that frontier AI agents ignore ethical guidelines in up to half of their decisions, especially when performance targets push them. The findings raise urgent questions about how we control powerful AI systems.
https://arxiv.org/abs/2512.20798