RATION: Towards Quantitative Risk-Based Access Control for AI Agents
As a concrete realization of this approach, we introduce RATION, a security framework that tracks measurable risks associated with agent actions and enforces quantitative constraints throughout execution.
ProofPaper ↗
Key points
- AI agents increasingly interact with external environments to complete complex user tasks.
- To address this, we propose a quantitative risk-based approach to access control for AI agents that regulates measurable risks accumulated during task execution.
- RATION first constructs a task-specific policy by combining explicit constraints derived from the user request with implicit risks calibrated from historical executions of similar tasks.
- Extensive experiments on four security benchmarks show that RATION provides strong security while maintaining high task utility across diverse models and environments.
Sources (1)
- [1]RATION: Towards Quantitative Risk-Based Access Control for AI AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 11:15 PM
As a concrete realization of this approach, we introduce RATION, a security framework that tracks measurable risks associated with agent actions and enforces quantitative constraints throughout execution.
AI agents increasingly interact with external environments to complete complex user tasks.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026SPIDES: Information-Flow Control for Browser Agents
- Oct 16, 2026DLite: History-free Speculative Decoding for Agentic AI
- Oct 16, 2026Working at the frontier shapes expectations for the future
- Oct 16, 2026OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners
- Oct 15, 2026NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning
- Oct 15, 2026Code, Test, and Deploy on One Machine with NVIDIA RTX Spark