[PDG 507] AgentSpec: Customizable Runtime Enforcement for Agents
Details
Link: https://arxiv.org/pdf/2503.18666
Title: AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM Agents
Track: Agents on Rails
Content: AgentSpec is a lightweight language for defining interpretable rules that enforce safety constraints on LLM agents at runtime, preventing over 90% of unsafe code executions and achieving complete safety compliance in the evaluated embodied-agent and autonomous-driving tasks, with millisecond overhead. LLMs can also generate these rules automatically, though their effectiveness varies and they miss some risks and violations.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.
In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.
