Skip to content

Start from a framework preset

Use a framework preset to start an Agent Intent Policy (AIP) from Intent Policies that OpenLatch has mapped to a standard framework, such as the EU AI Act or the OWASP Top 10 for LLM Applications. You need the editor role or higher. Nothing takes effect until you ratify it.

A framework preset is a built-in preset, read-only for every role. Each of its Intent Policies is a sentence OpenLatch drafted from the agent-execution content of the FINOS AI Governance Framework and adapted to its own wording. Framework preset sources and credit gives the credit, the license and the notice for each framework.

Framework preset
Agent Threat Rules
EU AI Act
FFIEC IT Handbook
IOSCO Supervisory Toolkit
ISO/IEC 42001:2023
NIST AI 600-1
NIST SP 800-53 Rev. 5
OWASP Machine Learning Security Top Ten
OWASP Top 10 for Agentic Applications
OWASP Top 10 for LLM Applications

These presets replace the earlier built-in presets OWASP Agentic Security and Testing verdicts. An AIP you already ratified from either keeps its sentences.

  1. Open Autonomy Zone, select the unit whose zone will hold the policy, open Policies and choose New intent policy.
  2. On the Brief step, choose Choose from a preset. Framework presets follow Examples and your organization’s presets, in alphabetical order. A card shows the framework’s title and a one-line description.
  3. Open a preset. Its Intent Policies are listed in a table you can search and filter, with the Framework items column naming the articles or controls each one cites. Nothing is selected when you open it. Select the Intent Policies you want, or all of the filtered ones with the header checkbox. Search matches a sentence and the items it cites.
  4. Choose Create AIP. The draft opens at the Policy step at once, with the test cases that ship with each sentence.

A framework item appears in the table only as the citation of an Intent Policy that serves it. An item that no Intent Policy serves is not listed, and OpenLatch shows no count of what the framework references.

Review a framework draft as you review any other: answer Yes or No for each Intent Policy on its own sentence, then ratify the whole draft. There is no bulk confirm. The sentences are written in English; you can edit one in your language during review, and that raises no flag.

On the How it runs step, Monitor stays preselected. For a draft made from a framework preset, the step adds this line:

Start in Monitor and move to Enforce as learning proposals come in. The choice is yours.

When your selection holds Intent Policies that ask a person, it also shows this one:

Some of these Intent Policies ask a person. By default the action is allowed and flagged for review, so the first occurrence is not prevented. The zone or the AIP can hold instead, on agents that support it.

A framework preset never sets how Ask behaves. Control levels lists, for each agent, how far an Intent Policy reaches.

After you ratify, open the AIP. Under each sentence from a framework preset, its framework items appear with no heading. If you edit a sentence, the items stay and the AIP says “This sentence was edited.”; restoring the original wording clears the line. Citations carry through edit drafts and restored versions. An AIP that did not come from a framework preset shows none.

The API returns the whole citation, including the framework entry each Intent Policy was written from. Framework preset sources and credit lists the fields.

Save a selection as an organization preset

Section titled “Save a selection as an organization preset”

Saving sentences from a framework preset as an organization preset keeps the sentences only. The dialog says so whenever a saved sentence has a citation: “The saved preset holds the sentences only. Citations stay with the framework preset.”

What a framework preset does and does not say

Section titled “What a framework preset does and does not say”

Each Intent Policy cites the framework items it relates to and may address only part of an item. A framework preset maps Intent Policies to its framework. It does not by itself establish compliance.

  • OpenLatch shows no score, percentage, badge, report or export for a framework.
  • An Intent Policy in Enforce takes effect as far as the agent’s control level and plane allow, as Control levels lists.
  • At a Block, the ratified sentence reaches the agent as the reason, and so reaches the model provider. What the agent receives on a Block says what is measured.
  • These Intent Policies govern the actions of an agent that misbehaves, not an agent built to get around them.

A framework preset can change over time. An AIP made earlier does not follow those changes: it keeps the Intent Policies and citations it was ratified with, and the product does not notify its owner. To use a preset’s current Intent Policies, create an AIP from it, or edit your AIP’s Intent Policies yourself.

A change reaches new drafts only. A draft you already opened keeps the Intent Policies it holds, and a ratified AIP works the same way when a rule or a whole preset leaves the catalog. A refresh follows the FINOS catalog, not the framework’s own text: a change a framework makes elsewhere reaches a preset only when the catalog carries it.