What an agent may do
Every proposed action is checked against the purpose, objective and intent the business approved. The answer is allow, approval required or deny.
Icebreaker · Runtime AI governance
Let agents act, without losing business control.
Identity establishes who an agent is. Access controls what it can reach. Icebreaker decides whether each proposed action is authorised, appropriate to its purpose and attributable to an accountable person.
What it controls
What Icebreaker controls, in business terms.
Without these controls, agents stay read-only or every action needs a person to check it, and the automation case does not pay back.
How far it can go alone
Limits are set per agent persona, so actions above a threshold go to a person before they run.
When a person decides
Actions that need approval go to the right supervisor with the recommendation and the context, so approval is faster.
That the approved action is the one executed
Your identity platform issues a one-time execution token bound to the approved action, so the agent cannot substitute another.
Who is accountable
Every governed action traces to an accountable person. Changes to agent personas are version controlled.
How to stop it
Revoke an agent or persona across the estate at once. If a check errors or anything tries to go around it, the action does not run.
The evidence. A record of each decision, with the reasoning behind it, held centrally for audit, risk and supervisors.
What changes
What governs an agent after access is granted?
Define authority
Register Purpose, Objective and permitted Intent.
Evaluate each action
Decide allow, approval required or deny.
Keep the evidence
Retain attributable decisions for investigation and oversight.
Runtime governance
Same access. Different authority.
Choose what an agent proposes to do.
Runtime decisionAllow
The action is within the approved business purpose and delegated limit.
Evidence recordPurpose → Objective → Intent → Action
Decision: Allow · Recorded to evidence storeSee it in practice
Follow a governed refund, step by step.
An agent investigates a customer complaint. Every proposed action is checked against its approved purpose before it runs.
Common questions
Questions about runtime governance.
Does Icebreaker replace IAM?
No. IAM determines identity and access entitlements. Icebreaker evaluates the proposed agent action against Purpose, Objective, Intent and Action, returning allow, approval required or deny.
Is Icebreaker patented?
Icebreaker is patent pending. Its maturity and supported integration patterns should be confirmed for the proposed deployment.
Next step
Assess one agent for controlled deployment Discuss runtime governance.
Tell us your current position and the result you need. We will discuss whether there is a practical, commercially sensible way to help.