SentinelleIA speaks the language of agentic AI security standards and covers the EU AI Act's compliance requirements.
MAESTRO is the Cloud Security Alliance threat model for agentic AI systems, published in July 2025. It decomposes an agentic system into seven layers, from the model itself up to the ecosystem of agents that call each other. Nine agents and a separate supervisor operate across all seven.
| Layer | MAESTRO domain | What the layer covers | SentinelleIA coverage |
|---|---|---|---|
| L1 | Foundation Models | Model level risks: jailbreaks, prompt level manipulation, unsafe or manipulated outputs. | Prompt Guard, LLM Security |
| L2 | Data Operations | Data pipelines, retrieval sources, third party datasets and model dependencies. | Supply Chain, Tool Protection |
| L3 | Agent Frameworks | The agent runtime itself: reasoning loop, tool invocation, mandate enforcement. | Supervisor (authorization boundary), AI Firewall |
| L4 | Deployment Infrastructure | Where agents run and how they reach the outside world. | Gateway, Tool Protection |
| L5 | Evaluation and Observability | Measurement, telemetry, detection of drift and of blind spots. | Visibility, metrics and alerting chain |
| L6 | Security and Compliance | Controls, evidence production, regulatory alignment. | Governance, SHA-256 chained audit log |
| L7 | Agent Ecosystem | Multi agent interaction, delegation, cascade failure between agents. | Resilience, supervisor |
Two components are cross cutting and therefore appear on more than one layer: tool protection on L2 and L4, and the supervisor on L3, L6 and L7. Every one of the nine agents is placed, none is orphaned. Covering a layer means having a control that operates there at runtime, not exhausting every risk the layer contains.
SentinelleIA aligns with the leading agentic AI security standards and the regulatory requirements in force.
The reference framework for risks specific to autonomous AI agents.
Application-level risks of large language models.
Layered threat modeling for agentic systems.
Adversarial tactics and techniques against ML systems.
Native compliance engine and enforceable proof for high-risk systems.
Security verification standard for AI systems.
AI risk management, GenAI and agentic profiles.
Security recommendations for generative and agentic AI.
AI management system.
SentinelleIA observes and controls agent interoperability protocols, and adds a sovereign proof layer on top.
Model Context Protocol. Interception and control of tool calls at runtime.
Agent-to-Agent. Visibility and control of exchanges between agents.
SHA-256 chained audit trail, enforceable proof native to the EU AI Act (art. 12). Designed and operated with no cloud dependency.
You submit your environment, we return an exposure report: measured attack rate and standards coverage.
Assess my exposure