When AI takes action, who remains accountable?
An approval given yesterday may not cover what an AI-assisted system does today. ManChine studies how authority, evidence and accountability hold up while systems act — and publishes open Runtime Governance standards so others can examine the work.
Start where your question is
Understand where an AI-assisted workflow can act without current authority
A proposed document-based review of one workflow: who approved it, whether that approval still holds, and what evidence shows the result.
See the proposed reviewRead the standards and the architecture
Published specifications and white papers, each with its version and status.
Open the libraryFollow RiCo implementation research and the PGC01 Guardian program
Early-stage development programs, described with their current stage and what they do not yet show.
See development programsRiCo — The Path to Runtime Integrity
What happens when permission changes after an action has already begun? Follow a fictional payment scenario to explore the gap between a local authorization check and a receiving system's later acceptance.
The film demonstrates a synthetic Python test only: no AI agent, real bank or real money. The proposed RiCo architecture has not been implemented or validated. Examples beyond finance are illustrative, not tested. Narration uses an AI-generated HeyGen voice.
RiCo — Too Early to Declare Success? Episode 2
When a workflow reports success, what evidence establishes what actually happened?
An educational film about the gap between a system reporting completion and an outcome being verified. Episode 2 is in review and has not been released.
Review status: publication pending. A viewing link will be added after release clearance.
The problem, in plain terms
An approval records intent.
It does not prove the authority still holds at the moment the action takes effect.
A log shows a system ran.
It does not show that the business outcome is the one that was authorized.
A recommendation is not an approval.
Even when, on screen, it looks like one.
Meet RiCo
RiCo is ManChine’s research into continuing runtime integrity: how a system can notice that the conditions behind an approval have changed while execution is active, and signal that a governance response is needed.
RiCo-01 — the responsibility
An architectural role in the Runtime Governance architecture, defined in ManChine’s publications.
RiCo — the research effort
ManChine’s implementation research toward that role. Public architecture documents and examples are available for review; a deployed enforcement system is not.

AI Workflow Boundary Review
ManChine’s proposed first advisory service: a document-based review of one AI-assisted workflow, from the event that starts it to the point where the action changes a business record.
What it is not
- A certification, audit opinion or legal opinion
- A security test or live-system examination
- A RiCo deployment, or a step toward buying one
- Proof that any control actually operates
How ManChine’s work fits together
Three separate tracks. Each is described with its own status, and none is a hidden requirement of another.
Advisory
Reviews an organization’s own documents and interviews. Does not touch systems.
Research & standards
Open, versioned publications anyone can read and challenge.
Development programs
RiCo implementation research and the PGC01 Guardian hardware program. Not offered commercially.
The review does not install, use or require RiCo.
How responsibility is divided
For technical reviewers: the four responsibilities defined in ManChine’s architecture publications, and how a material change leads back to a fresh determination.
- cGate — gives capability reach. It does not decide or enforce.
- RGE-01 — issues a fresh Admissibility Determination.
- REB-01 — enforces the Runtime Execution Boundary before consequential execution.
- RiCo-01 — addresses continuing runtime integrity once execution is active.
- When conditions change materially, RiCo-01 signals that a fresh governance response is needed.
What the public record shows — and what it does not
Every status on this site links to the record behind it.
RGE-01 remains an Architecture Review Draft.
See status of all workShown in public records
- Runtime Governance publications, listed with the identifiers and versions shown in the Research index
- RGE-01 Runtime Governance Evaluation, Version 1.2 — listed separately as an Architecture Review Draft, with formal proof closed: 35/35 frozen claims supported
- REB-01 proof scope recorded as closed (34/34) in the Research index
- Public RiCo architecture documents and examples
Not yet demonstrated
- Production runtime enforcement
- Prevention effectiveness in deployment
- Measured Guardian hardware validation, or an assembled prototype
- Production implementation or certification of any specification
- Delivered advisory engagements
Who is behind the work
Richard Colimon, Founder & CEO, ManChine AI Technology LLC.
ManChine’s publications are listed with identifiers, versions and status.
About ManChine