CONTROLLING AUTONOMOUS AI AGENTS : A Comparative Survey of Runtime Enforcement, Agent Architecture, Multi-Agent Monitoring and Collective Behaviour Detection
The rapid development of autonomous artificial intelligence agents raises a fundamental architectural question: How can human authority be preserved when AI systems increasingly plan, select, execute and adapt their own operational activities? Research addressing this challenge is developing across several distinct areas: runtime authorization, least-privilege access control, execution isolation, formal policy verification, system observability, anomaly detection and multi-agent coordination monitoring. This survey examines representative initiatives in these areas, with particular attention to thedistinction between controlling individual agent actions and identifying collective behaviour emerging through interactions among multiple agents. Nine principal research initiatives are examined: Faramesh, Tether, AgentGuardian,MCP-Secure, IsolateGPT, ACE, SentinelAgent, LumiMAS and Parity Swarm. A separate published systematic review of 58 primary studies is also considered asbackground evidence concerning the distribution of runtime-enforcement research. The survey then examines the ST-T1024 A1–A16 architecture and its complementary A17proposal for detecting emergent collective behaviour. The objective is not to establish the superiority of one architecture. It is to identify existingcontributions, examine their respective boundaries, expose unresolved research questions and establish a basis for independent technical comparison and experimentation. Keywords: Agentic AI; Autonomous AI Agents; Runtime Enforcement; AI Security;Multi-Agent Systems; Collective Behaviour; Swarm Detection; Runtime Monitoring;Deterministic Architecture; Spiking Neural Networks; ST-T1024; A17.
Authors
- M. TAALABI
- Team LLM-DIPLOMAT
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24411043
- Primary Topic
- Access Control and Trust
- Type
- article
- Field-Weighted Citation Impact
- 0.00