As LLMs are deployed as autonomous agents, safety failures increasingly involve consequential actions

As large language models (LLMs) are deployed as autonomous agents, safety failures increasingly involve consequential actions.

http://arxiv.org/abs/2608.27348v1