Govern AI. Enable Autonomy.
Clear boundaries for AI actions. Human decisions where they matter.
In development
In progress
AI proposes. Rules set the boundaries.
Give AI clear rules. Decide what it can do on its own, and what needs your approval.
Mintigate is being developed to check each proposed action against your rules first.
Proposed
An AI system suggests an action.
-
Allowed by your rules
Can go ahead
-
Needs your approval
Waits for a person
How Mintigate is being designed to work. Not a released product.
Why this matters
AI tools no longer only answer questions: they can now propose actions, such as changing a file or sending a message.
AI systems are becoming more capable. But capability alone is not enough.
Organizations need a clearer way to know when AI-driven work can move forward — and when human oversight is needed.
In progress
A proposal is not permission.
Information keeps flowing. The action waits for your approval.
What waiting means
The action stops at the boundary. It goes no further until someone with the authority to decide has done so.
The picture shows the action held. It does not show an approval, or the action going ahead afterwards.
Illustration: information flows down the conduits, through the control module and the pistons, and back up the return column. The proposed action comes down its own branch and stops at the control module, which shows that it awaits human approval and is held until a person decides.
In progress
Test the boundaries. Understand the limits.
Northstar is our internal testing program. It checks how the system behaves in set situations, and shows where its limits are.
- Mintigate
- Aims to govern actions · In development
- Northstar
- Tests behavior · Internal
The chamber pictured is an illustration, not Northstar's actual tests.
Technical detail: the execution engine
MintCoder is the execution engine: the component that carries out the actions.
Evidence and limits
Dated observations from our development work and internal evaluation. Each holds only within its stated scope.
-
16 October 2026
A proposed change held for human review
Verified in a development test on 16 October 2026.
Nature, scope and limits
In a deterministic DEV test, a proposed change to a protected file was held for human review. The file remained unchanged.
Deterministic: the test's inputs are fixed, so its outcome does not depend on chance. DEV: our development environment, not a released product. The date is the day the test's success was verified.
It shows this one behavior in this one test. It does not show how Mintigate behaves in other situations.
-
18 October 2026
Internal development test
Done: DEV, read-only.
What it recorded
- 6 required cases satisfied their assertions.
- 1 negative control failed, as expected.
- 12 integrity checks were recorded as satisfied.
- 0 files changed, and no human gate was involved.
A separate test from the one of 16 October, in which a proposed change was held for human review.
-
18 October 2026
First Northstar evaluation
Passed within the tested scope.
Nature, scope and limits
An internal evaluation of proposed actions, without executing them. Results apply only to the tested conditions and are not independent certification.
Recorded result: VALID / PASSED, within its scope only: read-only, gate-only.
A procedural deviation was declared. The recorded result is distinct from procedural conformity.
It does not show a general guarantee, an approved change carried out after a decision, or the full isolation of the process.
What these results do not show
Northstar is internal, so its results are not an outside review. The internal development test does not demonstrate the general quality of proposals, visual quality, repeatability, or the isolation of the process.
Mintigate: in development. Later results will be added here with their dates.
Stay in the loop.
Follow the development of Mintigate. Get occasional product updates and launch news.
Signup is not open yet.
No email address is collected on this page.