Assess action risk
Identify which agent actions require explicit authorisation and oversight.
Proprietary agentic AI controls
Control what agents can do.
Before they do it.
Our proprietary agentic AI control approach is designed to check authorisation before an AI agent performs an action that changes an enterprise system.
Our intellectual property
An execution-time
authorisation control.
An agent may be allowed to recommend a change without being allowed to execute it. Our approach puts a control between the proposed action and its execution.
Each permit is single-use, cryptographically bound to the exact action and verified at execution. The design addresses approval reuse, changes to approved actions, bypass of the approved route and duplicate execution.
From governance to a control requirement
- Define the permitted action
Describe exactly what the agent may change and under what conditions.
- Bind authorisation to that action
Require a valid, action-specific permit rather than a reusable approval.
- Verify at execution
Check the permit, hard rules and required data before the action occurs.
- Withhold execution when conditions fail
Fail closed where the underlying data is stale or incomplete.
Designed safeguards
Exact action.
Single use.
Hard rules.
The control design connects approval to the action being executed, with hard safety rules taking precedence over statistical models.
- 01
No reuse of prior approvals
A permit is designed for one authorised action and one use.
- 02
No silent change to the approved action
The permit is bound to the exact action, so changes invalidate that authorisation.
- 03
No model override of hard safety rules
Hard safety conditions remain decisive at execution.
- 04
No execution with stale or incomplete data
The control fails closed when required data conditions are not met.
Current status
Built and tested
at prototype stage.
A working prototype of each core component has been built and tested. A UK patent application is pending.
Prototype-stage capability
Discuss suitability, integration requirements and evaluation scope for your environment. Production deployment, scalability and independent validation would require further assessment.
How we can work with you
Define control requirements
Map permissions, safety conditions and evidence needs to your systems.
Scope an evaluation
Agree the environment, success criteria and evaluation of the approach.
Common questions
A clear basis for engagement.
Is this an off-the-shelf production product?
The current capability is a working prototype. Any evaluation or integration is scoped around the organisation’s systems, requirements and environment.
Is the patent granted?
No. A UK patent application has been filed and is currently pending. The control is described here at a high level.
Can this replace AI governance?
No. Runtime authorisation is one part of a wider system of governance, oversight, monitoring and assurance. Policies and accountability remain necessary.
Bring governance to the point of action.
Discuss the risks of agents acting on your enterprise systems and the controls those actions require.