Identify, Classify, Act: Per-Instruction Controlled Compliance for LLMs over External Data
Per-instruction compliance classification lets LLMs follow legitimate directives while rejecting injections in mixed-trust documents — a practical defence for RAG and tool-augmented deployments.
Summary written by editorial AI · Source link below
arXiv:2610.22874v1 Announce Type: cross Abstract: LLMs over external data face a per-instruction tension: a single document or tool return often interleaves legitimate directives with adversarial prompt injections, and the model must follow the former while rejecting the latter. Existing defenses suppress or trust all external instructions, collapsing the safety-utility frontier and forcing a choice between over-refusal and over-compliance. We argue the missing capability is \textbf{per-instruc
Editorial Analysis
As enterprises deploy RAG-based LLM applications over external data, per-instruction injection defence is critical to prevent data exfiltration and unauthorised actions.
Evaluate per-instruction compliance controls for any LLM application that ingests external documents or tool outputs.
Forward-looking interpretation drafted by editorial AI under human review — not a reproduction of the source. See methodology.
External link — opens at arXiv Crypto & Security in a new tab.
More from the AI Security Desk
- OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication1d
- Exposed Infrastructure Reveals How Ransomware Affiliates Used AI Agents to Execute Intrusions3d
- You gave your AI agent access, but who is giving it orders?3d
- AI Agents as Insider Threats: Inside Microsoft's Bet on a Unified Security Stack3d
- Critical x402 SDK Flaw Could Let Malicious APIs Drain AI Agents’ Token Balances3d