AIAdvanced#Architecture#Identity#Risk Management
Multi-Agent System (MAS)
M
Definition
A system in which multiple software agents or services perform bounded, coordinated tasks. Splitting work can improve separation of concerns, but it does not verify outputs, eliminate a single point of failure, or make a financial decision safe.
Source reviewed
- NIST: AI Agent Identity and Authorization opens in a new tab
Supports: NIST identifies service identity, authorization, auditability, and constrained permissions as key considerations for systems that use software and AI agents.
- OWASP: LLM06 Excessive Agency opens in a new tab
Supports: OWASP documents damaging outcomes when an agent-based application gives broad authority to unexpected, ambiguous, or manipulated model output.
Concept Diagram
Key Takeaways
1
MAS coordinates bounded tasks across agents; it does not establish better reasoning, independent verification, or trading safety.
2
Give every agent a named operator, identity, source and authority limits, schemas, data retention rule, time and cost limits, failure behavior, and revocation path.
3
Authenticate and log handoffs, allowlist destinations, make requests idempotent, and validate outputs independently before relying on them.
4
Shared agents can share outages, poisoned data, configuration errors, and prompt attacks; preserve disagreement and use safe fallbacks.
Practical Example
A collection service records a protocol notice with its source and timestamp, an extraction service creates a draft with citations, and a verifier flags missing evidence. The interface shows the disagreement to a reviewer; no component has access to a wallet or order API.
Finished learning?
Related Terms
AI
AgenticityAn informal way to describe how much an AI system can observe, plan, call tools, and act. It is not a standardized maturity score, a measure of trustworthiness, or permission to give an agent broad authority.
AI
Perception LayerThe part of an AI workflow that collects and labels inputs such as blockchain records, market data, and public communications. It supplies context for review; it does not establish truth, identify people, predict prices, or justify an action.
AI
Execution LayerA controlled component that prepares, validates, signs, submits, and monitors a state-changing operation. In an agent workflow, it must treat model output as an untrusted proposal, not as authority to move funds or call a contract.
AI
Agentic MeshAn architecture pattern in which multiple software agents exchange messages or delegate bounded tasks. It may distribute work, but it does not create collective intelligence, verify another agent's output, or authorize a financial action.
Related Deep Dives
Master the hub of AI
Explore all our strategic guides about AI to take your operations to the next level.
View all articles