The World Economic Forum Described the Agent We're Building

Mnemom Research | February 2026

---

In November 2025, the World Economic Forum and Capgemini published AI Agents in Action: Foundations for Evaluation and Governance — the most comprehensive governance framework for autonomous AI agents from a major international body to date. The report, developed with contributors from OpenAI, Google, Microsoft, Meta, AWS, Salesforce, and dozens of other organizations through the AI Governance Alliance, proposes a structured approach to classifying, evaluating, assessing risk, and governing AI agents as they move from prototypes to production.

At the center of their framework is the agent card — a structured description of an agent's capabilities, behavior, and operational context, inspired by Mitchell et al.'s Model Cards for Model Reporting (2019). The WEF proposes seven classification dimensions, nine baseline governance mechanisms, a five-step risk assessment lifecycle, and a progressive governance model that scales oversight with agent capability.

We've published a comprehensive mapping showing how the Agent Alignment Protocol (AAP) and Agent Integrity Protocol (AIP) implement every major recommendation in the WEF framework. The full mapping document is available on our research page and GitHub.

Here's what matters.

Describes vs. Binds

The WEF agent card is a classification document. It tells you what an agent does, how autonomous it is, and what environment it operates in — seven dimensions on sliding scales. This is useful. It gives organizations a common vocabulary for talking about their agents.

The AAP Alignment Card is a behavioral contract. It doesn't just describe what an agent does — it declares what the agent is permitted to do, what it must never do, when it must stop and ask a human, and how much it can spend autonomously. Every one of those declarations is machine-readable, schema-validated, and verifiable against observed behavior after the fact.

The WEF tells organizations what questions to ask about their agents. AAP provides the infrastructure for agents to answer them — verifiably. AIP provides continuous assurance that those answers remain true at runtime.

Coverage

The mapping addresses all four WEF pillars:

Classification. All seven WEF dimensions map to specific Alignment Card fields. Function maps to bounded_actions and forbidden_actions. Autonomy decomposes into what the agent can do, what it can't, when it must escalate, and how much it can spend. The WEF's HITL and HOTL governance models map directly to the principal relationship field that determines agent behavior at runtime.

Evaluation. The WEF calls for contextualized, multidimensional, temporal evaluation. AP-Traces record what the agent considered and chose in production conditions — not lab conditions. AIP Integrity Checkpoints analyze the agent's reasoning between every turn. Drift detection surfaces when behavior changes over time. These aren't evaluation recommendations — they're evaluation infrastructure.

Risk Assessment. The WEF's five-step risk lifecycle (define context, identify risks, analyze, evaluate, manage) maps to AAP's typed violation system. Each violation type carries a severity level that doubles as a risk ranking. A FORBIDDEN_ACTION is CRITICAL. A CARD_EXPIRED is HIGH. An UNDECLARED_VALUE is MEDIUM. The risk taxonomy is built into the protocol.

Progressive Governance. The WEF's nine baseline governance mechanisms — access control, monitoring, human oversight, traceability, trustworthiness, and the rest — each map to specific AAP/AIP implementations. This isn't aspirational. These mechanisms ship in the SDKs today.

Multi-Agent Risks

The WEF's forward-looking section on multi-agent ecosystems is where the mapping gets most interesting. The report identifies five emerging failure modes: orchestration drift, semantic misalignment, security and trust gaps, interconnectedness and cascading effects, and systemic complexity. These are real risks that will define the next phase of agent deployment.

AAP's Value Coherence Handshake addresses the first three directly — agents exchange Alignment Cards and verify behavioral compatibility before coordination begins. AIP's drift detection and cross-agent trace correlation address the last two. The WEF also envisions "governor" or "auditor" agents for scalable oversight. AIP's daimonion — the integrity analysis service that monitors an agent's thinking — implements this concept at the protocol level.

Why This Matters

The WEF framework joins the NIST NCCoE concept paper on agent identity and authorization, the EU AI Act's Article 50 transparency requirements, and the OECD AI Principles in forming a converging international consensus: autonomous agents need structured identity, verifiable behavior, and proportional governance.

AAP and AIP were designed for exactly this convergence. The Alignment Card is simultaneously an A2A agent card extension, an EU AI Act compliance artifact, a NIST-aligned identity document, and — as this mapping shows — a WEF agent card implementation. One artifact, multiple regulatory and governance frameworks satisfied.

The full mapping is available on our research page. The protocols are open source on GitHub.

---

Mnemom builds alignment and integrity infrastructure for autonomous agents. AAP and AIP are open source and available on npm and PyPI.