How to read this page
- Every step is one screen, or one moment, in a real run. The label under the title names where it happens, and the step reads in the same order each time: what is happening, the evidence, and then what to look for in it.
- Screenshots are real pictures of the Munitas console, the web app that people open in a browser. Click any picture to enlarge it.
- Every name of a person, patient or card holder in the example data is invented, and every screenshot is unedited apart from being cropped to the part that matters.
Words used on this page
- Organisation
- A company or hospital group that uses Munitas. Each organisation sees only its own data.
- Agent
- A program registered on Munitas, with a named owner, that Munitas runs on somebody’s behalf instead of a person reading data directly.
- Risk level
- One of the levels an organisation writes for how carefully it holds its agents, from the least strict to the most strict. A level says whether a change that lets an agent do more needs a second approver, how often the agent must be confirmed as still needed, and whether an agent at that level may be deployed at all. The person who registers an agent proposes a level, and the AI risk owner confirms it.
- Widening
- A change that lets an agent do more than the version before it: a new tool, a new external host, more data, or a different model. A widening has to be approved before the version can be deployed.
- Sponsor
- The person who answers for an agent. The sponsor is an approver of the department that owns it, and has to confirm from time to time that the agent is still needed and held at the right level.
- Agent registry
- A list of every agent in the organisation, open to everybody signed in. It shows the department that owns each agent, its risk level, its sponsor and what needs attention.
- AI risk owner
- The one person who answers for how the organisation holds its AI agents. The role writes the risk levels and the approved tools list, confirms the level of each agent, and gives the second approval on a change to an agent where its level asks for one.
- De-identification reviewer
- A role that allows a person to review data from which identifying details have been removed, and to decide whether it may be released.
- Deploy
- To make one version of an agent the one that runs when somebody starts a run. Only one version of an agent is active at a time.
- Department
- A team inside an organisation that owns some of its data, such as Cardiology.
- Version
- One sealed state of a dataset. Changing the contents means making a new version, never editing an old one.
- Version of an agent
- One uploaded and sealed copy of an agent’s code. Uploading again makes a new version, and the older ones stay.
The organisation writes down how carefully each kind of agent is held
An AI agent is a program that works on data on somebody's behalf. Some agents only tidy appointment reminders. Others write notes about patients. An organisation does not want to decide how careful to be one agent at a time, so it writes a short list of risk levels, from the least strict to the most strict, and every agent is placed on it.
Okafor is the AI risk owner of the Health organisation, the one person who answers for how the organisation holds its AI agents. Writing the list of levels belongs to that role. The person who registers an agent proposes a level for it, and the AI risk owner confirms it. Each level says three things: whether a change that lets an agent do more needs a second approver, how often somebody has to confirm that the agent is still needed, and whether an agent at that level may be deployed at all. To deploy an agent version is to make it the one that runs. The list in force is version 1, written when the organisation first held its agents to levels. Every later change adds a version, and every earlier version is kept.
Each level has a name, a label for the law or standard it corresponds to, a sentence on what an agent at that level is like, and the number of days after which the agent has to be confirmed again. Two choices complete it: a second approver is needed to widen an agent at that level, and an agent at that level may not be deployed. A widening is a change that lets an agent do more than the version before it. The level called Not allowed has the second choice ticked, so no version of an agent at that level can be deployed.
A change is saved as a new version, and the old one is kept
A list of levels is never edited in place. Every save adds a new version, with the name of the person who made it, the date and time, and a written reason.
Agents at Moderate were going a long time between checks. Re-confirming means that somebody checks that the agent is still needed and still held at the right level. So Okafor shortens the time for Moderate from 180 days to 120. Only this one number changes. The other levels stay as they were.
Further down the same form, Okafor writes the reason for the change. The platform will not save a list without one. The switch that holds agents to the levels stays ticked: an agent then needs a confirmed level before it can start a run, and a version that widens an agent has to be approved before it is deployed.
The page now shows the new version as the list in force. The reason sits next to the name and the date, so that a reader who was not there can see why the organisation changed its rule. Only the Moderate row differs from before.
A decision made earlier was judged under the list in force at the time. Because every version stays, an auditor can always say which rule applied on a given day. The page leaves out the version in force, which is shown above, and lists the one before it, the first list. The history shows what the rule was before the change.
Who may change the list, and what a change does
The levels decide how strictly agents are held, so the people who judge agents must not be able to relax the rules that apply to them.
Imani is a de-identification reviewer in the Health organisation, a person who reviews data from which identifying details have been removed, and who may also check what an agent produced. Every person signed in can read the list, so that anybody can see what an agent is held to. Only the AI risk owner can set it, and the page says so in place of a button.
Okafor follows the agent care-gap-finder from the Agent registry. It belongs to the Oncology department and is held at Moderate, and its sponsor, the person who answers for it, is Mensah, an approver for Oncology. An agent keeps the level it was confirmed at, but how often it has to be confirmed again follows the list in force. So the page now says every 120 days, and the date by which it is due again is 60 days earlier than it would have been under 180. The card lower down is the record of the agent's first version, approved when the agent was set up: it names who approved for the department and who approved for the AI risk owner.
