Jakten Risk AI Data Handling

Audience: Company administrator, IT and security reviewers

Last updated: 2026-10-16

Applies to: Jakten Risk with a Jakten AI subscription, under the *Jakten AI Data Processing Terms* version 1.2

This page summarizes what Risk's AI features send, where it goes, and what is kept. The *Jakten AI Data Processing Terms* (https://jakten.net/legal/ai_data_processing/v1.2), Schedule D, are the full list and govern where the two differ.

1. Summary

AI is off until a company administrator opts in to the *Jakten AI Data Processing Terms*, and the company has a Jakten AI subscription.
AI drafts; a person decides and enters every number. A drafted risk event never carries a probability, cost impact or schedule impact: the person accepting it enters them, and nothing drafted enters a simulation until it is accepted. A drafted summary prints on the report only when a person accepts it.
AI never touches the simulation. It never runs inside a simulation, never changes its inputs or seeds, and never changes a sealed result.
Risk works without AI. The risk model, simulations, results, seals, the inbox of Quote's risks and every report work with AI off and with no internet.

2. What each feature sends

Feature	What it sends
Draft summary	The sealed result's figures and the drivers of its contingency (such as estimate uncertainty, quantity growth and risk events), as the seal froze them
Draft risk events	The scenario's estimate, by WBS element and cost type, with each element's description (the largest 400 lines; for the scenario bound to an estimate Version, as Jakten Assurance published it); the Knowledge Assurance published for that Version (each item's type, title, statement, rationale and category), when the scenario is bound to that Version; the risks recorded in the Jakten Quote awards bound to the project that were accepted or are awaiting clarification and that the bidder does not carry (each one's title, type, severity, and exposure where it was priced, with its currency); the names and WBS elements of the risk model's events; and the names of the project's drafted events still awaiting a decision, so none is drafted twice

Feature

What it sends

Draft summary

The sealed result's figures and the drivers of its contingency (such as estimate uncertainty, quantity growth and risk events), as the seal froze them

Draft risk events

The scenario's estimate, by WBS element and cost type, with each element's description (the largest 400 lines; for the scenario bound to an estimate Version, as Jakten Assurance published it); the Knowledge Assurance published for that Version (each item's type, title, statement, rationale and category), when the scenario is bound to that Version; the risks recorded in the Jakten Quote awards bound to the project that were accepted or are awaiting clarification and that the bidder does not carry (each one's title, type, severity, and exposure where it was priced, with its currency); the names and WBS elements of the risk model's events; and the names of the project's drafted events still awaiting a decision, so none is drafted twice

Dates, and the names Risk records for owners, are removed from the text Risk holds (estimate descriptions and event names) before a draft of risk events sends it. The Knowledge and Quote's risks are sent as Assurance and Quote published them, including any date or name a person wrote in them.

Never sent: user names and user identifiers, and the names Risk records for owners; recorded actions and their due dates; the scenario's name; and any other project's data. Draft summary never sends the working risk model, other scenarios, or any run but the sealed result it summarizes. Draft risk events never sends the model's probabilities, impacts, delays or ranges, its other settings, or any simulation result.

3. The route

Your Platform Server (or, in a single-user deployment, your workstation) sends each AI request to Jakten's gateway at licensing.jakten.net, the same host already used for license activation and validation. The gateway forwards it to Anthropic, PBC, the provider of the Claude models, and returns the answer.

Jakten holds the provider account. Your deployment holds no AI provider credentials, and you have no separate agreement with Anthropic. Anthropic is a subprocessor of Jakten for this service.
Requests identify the person who used a feature only by a coded reference, never by name or email address.

4. What is kept

By Jakten's gateway: a usage record of each request (your company and installation, the feature, the model, the tokens, the credits, the time, the coded reference to the person, and the AI provider's reference for the request), and the provider's error message when it returns an error. It does not store what was sent or the answers.
In your own Jakten database:
every drafted analysis summary, with the text drafted, the check's result, the model, who drafted it, and who accepted, edited, rejected or removed it. Only an accepted summary prints, on the Full Report of the sealed result it was drafted from;
each drafted risk event, with its name, notes, WBS element, its sources, the published records it was drafted from, and who accepted or rejected it.
At Anthropic: Anthropic states that it does not use inputs or outputs of its commercial API to train its models by default, and deletes them within 30 days, except content it must keep by law or flags under its Usage Policy (kept up to 2 years). Jakten does not opt your data in to training. The terms, section 5, give the dates these statements were confirmed.

5. Turning AI on and off

To turn AI on, a company administrator opens Administration → Licensing → Subscription and billing and, in the AI panel under AI data processing, selects Opt in…, reads the *Jakten AI Data Processing Terms* it links, and confirms. The opt-in covers the whole company and every Jakten module. Your company also needs a Jakten AI subscription (Company Administrator Guide, section 10).

To turn AI off, an administrator selects Withdraw… in the same place. Jakten's gateway then refuses AI requests for your company, so "off" does not depend on each workstation. A new version of the terms pauses AI until an administrator opts in to it.

Withdrawing deletes nothing already in your Risk database: accepted events, summaries and decisions stay yours.

6. What still works without AI or without internet

Works with AI off or offline	Needs AI and internet
The risk model, simulations, results and seals	—
The Full Report, without an analysis summary	Draft summary
Adding risk events by hand, and the inbox of Quote's risks, with every accept	Draft risk events

When an AI feature cannot run, Risk shows why: offline, AI off, not opted in, no subscription, or credits used up. It never skips the step silently.