Book a demo
Topics · GOVERNANCE-AGENTS

Payments Agent Registry Stop Control | Moring

Stop control is the guaranteed ability to halt a payment agent's actions through the registry: state suspension honored by the runtime at every consequential call, with the halt attributed, evidenced, and scoped to the offending agent. In payments, where agents move funds, stop control is the difference between an incident and a loss event.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading

Short answer

Stop control is the guaranteed ability to halt a payment agent's actions through the registry: state suspension honored by the runtime at every consequential call, with the halt attributed, evidenced, and scoped to the offending agent. In payments, where agents move funds, stop control is the difference between an incident and a loss event.

How it works

Stop control answers the question that becomes urgent the moment agents can move money: commentary on the agentic shift notes that AI agents are writing code, moving money, and making decisions with real consequences while most organizations have no way to stop them from going too far. Payment flows sharpen the concern because a payment released by a misbehaving agent settles whether or not anyone intended it. Stop control is therefore a designed property of a payments agent registry, not an aspiration bolted on after deployment. Sources: https://www.facebook.com/quartznews/posts/new-tools-want-to-stop-ai-agents-from-going-rogue-with-your-money-and-codeas-ai/1473377064658107/ https://www.sweet.security/agent-security/what-is-ai-agent-governance

The mechanism is registry state evaluated at execution time. Each payment agent enrollment carries a state, and the runtime checks it before every consequential call, so suspending the agent denies the next payment execution within seconds. Because the same registry that admits an agent also removes it, stop control inherits the enrollment's permissions and scopes, and the suspension event is logged with who triggered it, when, and which payment calls were denied. Sources: https://www.moring.ai/ai-governance/ https://www.sweet.security/agent-security/what-is-ai-agent-governance

Stop control must also cover the override case. The mirror image of a runaway agent is an agent that blocks a legitimate payment with no path to release it, which is why governed patterns gate consequential actions to a named human rather than leaving the decision with the agent. A complete stop-control design pairs agent suspension with a human override path recorded in the registry, so an accountable approver can release a held payment and the release is sealed as evidence alongside the suspension. Sources: https://www.moring.ai/ai-governance/ https://www.sweet.security/agent-security/what-is-ai-agent-governance

Visibility is the precondition for stopping anything. A registry giving complete visibility, control, and oversight into every agent deployed across the enterprise, with compliance automation and risk surfacing, is the inventory that makes targeted stops possible; without it, responders face shared service accounts and embedded tokens that identify no one and cannot be revoked safely. New products emerging to enforce human-set rules before agent actions go through, particularly around money and code, reflect the same market recognition that stopping agents is now a category. Sources: https://www.credo.ai/ai-agent-registry https://www.facebook.com/quartznews/posts/new-tools-want-to-stop-ai-agents-from-going-rogue-with-your-money-and-codeas-ai/1473377064658107/

Design properties for payment stop controls follow from money movement. Authority is explicit: named people can trigger the stop, and attempts are attributed. Scope is targeted: the offending agent is suspended, not the payment platform, so other flows keep working. Speed is bounded: containment lands within seconds, before the next settlement window. Recovery is gated: re-enabling requires owner and risk sign-off on a changed enrollment, and the whole sequence lives in the registry record. Sources: https://www.sweet.security/agent-security/what-is-ai-agent-governance https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/ai-agents/governance-security-across-organization

Finally, exercise the stop path on a schedule. Drills should suspend a payment agent in a controlled environment, verify the runtime denied the subsequent payment call, confirm the decision record sealed correctly, and rehearse the override path for a legitimately held payment. Governed patterns verify that each delegated action is expected, permitted, and reversible before an autonomous agent changes systems or data; a stop control that has never been exercised is a claim, and in payments claims do not reverse settlements. Sources: https://www.sweet.security/agent-security/what-is-ai-agent-governance https://www.dataiku.com/blog/ai-agent-governance-guide - federalreserve.gov: https://www.federalreserve.gov/supervisionreg/srletters/SR2602.htm - ithandbook.ffiec.gov: https://ithandbook.ffiec.gov/ - artificialintelligenceact.eu: https://artificialintelligenceact.eu/high-level-summary/ - eur-lex.europa.eu: https://eur-lex.europa.eu/eli/reg/2022/2554/oj

Frequently asked questions

What is stop control in a payments agent registry?

The guaranteed ability to halt a payment agent's actions: registry state suspension honored by the runtime before every consequential call, with the halt attributed to a named authority, logged with the payment calls denied, and scoped so only the offending agent stops. In payments it is the difference between an incident and a loss event, because a released payment settles regardless of intent, so the stop must be designed, not assumed.

How fast must a payment agent stop be?

Within seconds, before the next settlement window. The runtime checks registry state before each consequential call, so flipping the enrollment state denies the next payment execution immediately without infrastructure change or credential revocation. Bounded speed matters because machine-paced agents can generate payment volume no human cadence can intercept, and containment that lands after release converts a controllable event into an irreversible one.

Should stop control include a human override for held payments?

Yes, because agents also block legitimate payments. Commentary describes a fraud agent that blocked a genuine executive transfer with no override function, forcing an engineering escalation while a contract deadline passed. The complete design pairs agent suspension with a human override path recorded in the registry: an accountable approver releases the held payment, and the release is sealed into the same evidence, preserving accountability in both directions.

What makes a stop control auditable?

The registry record of the event: who triggered the stop, when, which payment calls were denied, what state the enrollment moved to, and who signed off on re-enabling. Because the registry is also the admission control, the stop sequence lives in the same system that governs the agent's authority, so auditors see a single consistent record rather than reconciling logs from separate systems after the fact.