Karsenics AI Governance and Acceptable Use Policy

The short version. KarsenicsOS includes an optional AI assistant called Milton AI. Milton AI reads, explains and drafts, but a person makes every decision and sends every message. No one trains an AI model on your organization's data. Your organization's Primary Administrator can turn Milton AI off at any time, and then nothing from your organization goes to the model provider.

This policy covers Milton AI, the AI assistant in KarsenicsOS. It applies to Karsenics PBC and to every client organization, and its staff and volunteers, that use KarsenicsOS. It is incorporated by reference into the Master Services Agreement. The Platform Acceptable Use Policy also applies.

1. Milton and Milton AI are different things

Milton is the name on KarsenicsOS's automated emails, such as confirmations, receipts and reminders. Milton does not use AI and sends nothing to any model provider. This policy is about Milton AI, the assistant in the administrative console.

2. What Milton AI does

2.1 Milton AI helps an organization's staff in four ways:

• it answers questions about KarsenicsOS and about the organization's own events, using a written knowledge base and the records the person asking is allowed to see;
• it drafts memo text that a staff member can edit, and then send or discard;
• it explains in plain words why the parade lineup tool placed an entry where it did, and why a spacing recommendation was made; and
• it can fill in a report for a staff member to review and save.

2.2 Memo drafts are built from summary figures, an audience label and a recipient count. Registrants' names and email addresses are not sent to the model to draft a memo.

2.3 Milton AI runs on Claude, a model made by Anthropic, PBC. Anthropic is on the Karsenics subprocessor list at karsenics.com/trust.

3. A person decides

3.1 Milton AI does not, on its own, change a setting, set a price, approve or reject an application, assign or move a booth or lineup spot, issue or deny a refund, accept or reject a compliance document, suspend or remove a participant, or send a message.

3.2 Anything Milton AI drafts or fills in takes effect only when a staff member reviews it and chooses to send or save it. When Milton AI can't answer a question, it can pass the question to a person the staff member chooses, by email, only after the staff member confirms.

3.3 AI may draft, suggest, summarize and explain. A person decides.

4. Turning Milton AI off

4.1 The organization's Primary Administrator can turn Milton AI off at any time in the administrative console, and turn it back on later.

4.2 While it is off, no one can use Milton AI for that organization, Karsenics staff included, and nothing from the organization is sent to Anthropic. Reports, automated emails and the rest of KarsenicsOS keep working.

4.3 Karsenics can also switch Milton AI off for an organization's site.

5. Your organization's data

5.1 No training. Participant data and organization content are never used to train, fine-tune or improve any AI model, whether Karsenics' or anyone else's. Anthropic's commercial terms do not allow it to train its models on the content Milton AI sends.

5.2 Where it goes. Anthropic stores data in the United States. Its model may answer from data centers in other countries.

5.3 How long it is kept. Anthropic handles that content under its commercial terms and keeps it only as long as those terms allow.

5.4 Ownership. Your organization owns what it puts into Milton AI and what Milton AI produces from it. Karsenics claims no rights in either.

5.5 Limits and records. Each organization has a monthly limit on Milton AI answers and drafts. Every question asked of Milton AI is recorded in a question log.

6. AI answers can be wrong

Milton AI's answers, drafts and explanations can be wrong, incomplete or misleading. Karsenics does not promise that they are accurate. Check anything before you rely on it, publish it or send it.

7. What Karsenics will not build

Karsenics will not build into KarsenicsOS:

• automated scoring of applications or jury submissions that ranks or rejects people without a person deciding;
• any guess about a participant's identity, sexual orientation, gender identity, health, immigration status or any other protected characteristic, from any data, for any purpose;
• facial recognition or any other biometric identification;
• predictive scores attached to a person's record, such as a likelihood of not showing up;
• emotion or sentiment analysis of individuals; or
• any feature that sends participant data to a model provider that trains on it.

8. Testing for fairness

Karsenics serves Pride organizations and community festivals, whose participants are often people that general AI models have handled badly. Before Karsenics ships any AI feature that scores, ranks or reviews applications or jury submissions, it will test whether the feature treats people differently based on names, organization names, or language tied to those communities. A feature that does will not ship.

9. New AI providers

Before any new model provider receives organization data, Karsenics adds it to the subprocessor list and gives client organizations at least 30 days' notice.

10. What organizations and their staff must not do

When using Milton AI, you will not:

• create unlawful content, or content that breaks the Platform Acceptable Use Policy;
• impersonate a real person, or present AI writing as a person's own where that would mislead;
• enter passwords, full card or bank account numbers, or government identification numbers; or
• publish or send AI output without reading it first.

11. Questions and reports

Questions about this policy, or a Milton AI answer that looks wrong or harmful: legal@karsenics.com. Privacy questions: privacy@karsenics.com.

12. Changes

When this policy changes, the new version is posted here with a new date, and each earlier version stays available at its own dated address. If a change reduces what this policy promises about an organization's data, Karsenics will tell each organization's Primary Administrator at least 30 days before the change takes effect.