Constellation vs AI Governance Platforms

AI governance platforms are model-centric. They:

• •Monitor ML model performance, drift, and degradation
• •Detect and measure algorithmic bias across protected categories
• •Assess AI risk against regulatory frameworks (EU AI Act, NIST AI RMF)
• •Generate model cards and explainability reports
• •Track model lineage, versioning, and deployment inventories
• •Manage AI policy documents and attestation workflows

This is valuable work. If you deploy machine learning models, you need tooling like this. The question is whether this is the same problem as institutional governance.

AI governance platforms ask: “Is this model behaving correctly?” Constellation asks: “Is this action legitimate given what the organisation has decided?”

Consider a scenario: your AI agent approves a $200,000 vendor contract.

The model behaved perfectly. The action was still institutionally illegitimate. AI governance platforms would see nothing wrong. Constellation would catch it.

AI governance platforms are designed for model oversight. They cannot:

• •Evaluate whether an AI agent’s action conflicts with a board resolution or institutional commitment
• •Enforce spending thresholds, approval sequences, or delegation boundaries in real-time
• •Intercept MCP tool calls at the moment of execution and check them against institutional constraints
• •Route escalations to the appropriate human authority with full decision context
• •Build institutional precedent from past governance decisions
• •Allow anyone in the organisation to challenge a constraint through a formal contestation process

These aren’t shortcomings. AI governance platforms weren’t built for institutional governance. They were built for model governance. The distinction matters.

AI governance platforms assume that if the model is performing well, the outputs are safe to act on. But institutional legitimacy is not a model performance metric.

An AI agent can produce a perfectly unbiased, high-confidence recommendation that violates an institutional commitment, exceeds delegated authority, or contradicts an active board resolution. No amount of model monitoring will catch this because the problem is not in the model — it’s in the gap between model output and institutional action.

This is the layer Constellation occupies. It sits between AI capability and institutional consequence — ensuring that what an AI agent can do is checked against what the organisation has decided it should do.

AI governance platforms and Constellation are complementary. AI governance ensures models behave well. Constellation ensures institutions act legitimately — regardless of whether the action was triggered by a human or an AI agent.

If you use AI agents that take consequential actions — approving expenditure, publishing communications, modifying contracts, triggering workflows — you need both layers. AI governance watches the model. Constellation governs the institution.