# Researchers Propose Blocking Agent Actions Unless Every Influencing Principal Is Authorized as Hyundai Tests NVIDIA's Agent Runtime

The Conflux paper replaces prompt-injection detection with a check on data provenance and existing access controls, and NVIDIA's Korean partners are building on its OpenShell runtime for agent security.

- Published: 2026-11-09T06:17:39.329Z
- Canonical: https://polylog.news/ai/2026-11-09/researchers-propose-blocking-agent-actions-unless-every-infl
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CR](https://arxiv.org/abs/2611.08768), [NVIDIA Blog](https://blogs.nvidia.com/blog/korea-secure-agentic-ai/)

Raya Buckley and Christian Schroeder de Witt posted Conflux on November 5. The paper defines privilege escalation as the security goal for LLM agents that combine inputs from several principals before acting. Its rule is that an action may…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-11-09/researchers-propose-blocking-agent-actions-unless-every-infl (subscription information: https://polylog.news/pricing).