ARES: Securing Agents for Computer Use Through Endpoint Resource Mediation and Behavioral Guardrails
Open Access
- 4 September 2026
- journal article
- research article
- Published by MDPI AG in Electronics
- Vol. 15 (17) , 4007
Abstract
Large language model (LLM)-based agents are evolving into agents for computer use (ACUs) that read files, invoke applications, communicate over networks, and operate graphical interfaces, moving the effective security boundary from model inputs and outputs to autonomous actions that alter endpoint state. Conventional identity and access controls remain applicable and necessary, but they are authorized based on identity, resource, and network policy rather than on the semantic scope of the active task or the provenance of the instruction that triggered an action. This paper presents ARES (Agent Resource Enforcement and Security), an action-centric framework that inserts enforceable authorization between agent-generated tool calls and protected resources, combining a Resource Proxy Layer for interception, a Behavioral Guardrail Engine for task- and context-aware authorization, and a Multi-Agent Trust Boundary Manager for provenance and taint propagation. We implement ARES-lite and evaluate it against file exfiltration through indirect prompt injection, prompt infection propagation, and internal-network access abuse. Under the controlled replay-based evaluation, integrated ARES-lite reduced the observed attack-success rate from 100% under the permissive baseline to 0% across the three evaluated scenarios, while preserving the predefined handling of eight benign and ambiguous tasks with no false positives. A closed-loop evaluation in which the agent re-plans after an intervention, an adversarial boundary-case suite covering path aliasing, address encoding, look-alike destinations, and taint laundering, and a comparison against representative prompt-filtering and tool-allowlist defenses further characterize the enforcement path; a model and temperature sweep show that baseline exposure varies with model capability, whereas the mediated outcome does not.Keywords
All Related Versions
This publication has 21 references indexed in Scilit:
- Cryptanalysis and Improvement of a Video Cryptosystem via Chaos and S-BoxACM Transactions on Multimedia Computing, Communications, and Applications, 2026
- Prompt Infection: LLM-to-LLM Prompt Injection within Multi-agent Systems2026
- A Comprehensive Survey of Agents for Computer Use: Foundations, Challenges, and Future DirectionsJournal of Artificial Intelligence Research, 2026
- Model Context Protocol (MCP): Landscape, Security Threats, and Future Research DirectionsACM Transactions on Software Engineering and Methodology, 2026
- LLM-Based Multi-agent Systems: Frameworks, Evaluation, Open Challenges, and Research Frontiers2026
- From prompt injections to protocol exploits: Threats in LLM-powered AI agents workflowsICT Express, 2025
- AgentSentinel: An End-to-End and Real-Time Security Defense Framework for Computer-Use Agents2025
- Multi-Agent Security Tax: Trading Off Security and Collaboration Capabilities in Multi-Agent SystemsProceedings of the AAAI Conference on Artificial Intelligence, 2025
- IsolateGPT: An Execution Isolation Architecture for LLM-Based Systems2025
- Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023