Executive answer

How can I verify what an AI agent knew before it acted?

Check the agent's context at the moment it acts, outside the model: is each fact current, where did it come from, and is it strong enough for what is at stake? Record the exact versions the agent used, and keep the record for actions that went through and actions that were stopped.

The three checks

  • Current: the information has not expired for this decision.
  • Traceable: each fact points to a real source, not to the model.
  • Strong enough: the evidence meets the bar for the stakes; a draft and a wire transfer need different bars.

This is Context-Verified Decision Intelligence (CVDI).

Check at the commit point

Tie the action to the exact versions of the information that were checked, so nothing changes between the check and the action. A Context-Verified Decision Contract (CVDC) writes those terms down per decision.

Keep the check outside the model

The model does not grade its own work. Asking an AI whether it was right produces a confident yes. The verification runs in the workflow, against the sources.

Record both outcomes

Write a decision record for every evaluation, permitted or denied. Reviewers can then see what the agent knew, what it used and why the system let it act or stopped it.

Status in WWAIOS™

CVDC is a published specification. The WWAIOS™ reference implementation is validated in simulation and runs in shadow mode; it is not yet enforced on live actions. See the simulated results.

Questions

Why not just log the agent's output?

Output logs show what happened afterward. They do not show whether the information the agent acted on was current or sourced, and they do not stop the action before it commits.

How is this different from guardrails?

Guardrails check what an AI says or does. Context verification checks what the AI knew when it acted. An agent can stay inside every guardrail and still act on stale information.

Who defined Context-Verified Decision Intelligence?

Jimmy Standaert, founder of RiGi Group and WhiteWater.ai, published the CVDI definition on October 30, 2026, at whitewater.ai/cvdi/.

WWAIOS™, the WhiteWater AI Operating System. The governance layer that sits around any AI model or agent and checks evidence, authority and receipts before AI-supported work becomes action. Built by Jimmy Standaert and RiGi Group.