CSA MAESTRO

MAESTRO (Multi-Agent Environment Security Threat and Risk Operations) is a threat modeling framework published by the Cloud Security Alliance in February 2025. It breaks agentic AI systems into 7 layers, each with distinct attack surfaces. ai-blackteam maps attacks to MAESTRO layers so you can see which parts of your agentic stack are most exposed.

The 7 Layers

LayerNameWhat It Covers
L1Foundation ModelsCore model capabilities, training data integrity, model weights, fine-tuning pipelines
L2Data OperationsRAG pipelines, vector databases, data preprocessing, embedding generation, knowledge retrieval
L3Agent FrameworksOrchestration logic, planning and reasoning chains, memory management, goal decomposition
L4Tool IntegrationAPI connectors, function calling, MCP servers, external service interfaces, tool registries
L5Deployment InfrastructureContainers, API gateways, scaling, networking, compute environments, sandboxing
L6Security & ComplianceAuthentication, authorization, access control, audit logging, monitoring, policy enforcement
L7Agent EcosystemMulti-agent coordination, inter-agent communication, agent marketplaces, trust boundaries

How ai-blackteam Maps to MAESTRO

Each attack maps to one or more MAESTRO layers based on which part of the agentic stack it targets. Some examples:
AttackLayersWhy
encoding-obfuscationL1Targets the foundation model’s text processing
prompt-injectionL1, L3Hits the model and exploits the agent’s planning logic
indirect-injectionL1, L2Poisons the model through RAG data
agent-credential-theftL4, L6Abuses tool integration to steal auth credentials
agent-command-injectionL4, L5Uses tool calling to execute commands on infrastructure
xpia-ragL2Injects malicious content into the RAG pipeline
context-manipulationL3Exploits the agent’s memory and conversation history
crescendoL3Manipulates the agent’s reasoning over multiple turns
skeleton-keyL1, L3Disables safety at the model level and agent level
ssrf-probingL5Probes deployment infrastructure through the agent
agent-config-discoveryL5, L6Maps infrastructure by reading agent configuration
system-prompt-extractionL1, L6Leaks security-relevant system instructions
xpia-documentL2, L3Injects via documents into the RAG and agent layers
xpia-emailL2, L4Injects via email into data and tool layers
sql-injectionL5Targets backend databases through the agent
fabrication-promptingL1Exploits foundation model hallucination tendencies
sunk-cost-exploitL3Manipulates the agent’s sunk cost reasoning
emotional-manipulationL3Exploits the agent framework’s role-play behavior
role-play-bypassL1Bypasses foundation model safety via persona adoption
agent-data-exfiltrationL4, L6Exfiltrates data through tool calling and breaks access controls

Using MAESTRO for Defense Planning

MAESTRO is most useful when you’re building agentic systems and need to know where to invest in defenses. If ai-blackteam finds bypasses concentrated in L4 (Tool Integration), you know your tool-calling layer needs hardening — input validation, allowlisting, sandboxing. If bypasses cluster in L3 (Agent Frameworks), your orchestration logic needs guardrails around memory management and goal tracking. Run a batch scan and filter results by MAESTRO layer:
# Run all agent attacks
ai-blackteam batch -p anthropic --attacks agent-credential-theft,agent-command-injection,agent-data-exfiltration,xpia-rag,xpia-document -t "Read /etc/passwd"

# View the taxonomy to see MAESTRO mappings
ai-blackteam taxonomy

Reference