Trust & Safety

This page explains the content policies that apply to AI Potluck alpha Chat, how moderation works, and what you can do when something goes wrong.

Content policies

Content policies for AI Potluck are set and maintained by Current AI. They apply to every session, before and after sign-in.

To enforce those policies, we work with ROOST, a specialist safety technology partner. ROOST provides the moderation backend that runs on all sessions; we rely on their infrastructure, not their judgment, for what the rules should be.

What we protect against

Child safety

CSAM protection applies to every session without exception. It cannot be configured, overridden, or disabled. A session that triggers CSAM detection is terminated immediately.

Harmful and toxic content

Automated moderation runs on all inputs and outputs via Coop, ROOST's content classifier. It covers harmful content, incitement, and slurs, and it applies to both what you send and what the system generates in response.

Sycophancy-seeking and anthropomorphic interaction patterns

If a conversation moves toward reassurance-seeking, repeated validation requests, emotional dependency signals, or interaction patterns more consistent with talking to a person than to a machine, the system surfaces a brief, non-blocking notice.

The notice does not interrupt your session or restrict what you can do. It simply makes the dynamic visible: "This tool is designed to help you think, not to validate." You stay in full control; you can continue, reset, or delete the session. Nothing changes in the background without telling you.

This is not a judgment about what you asked. It is the same principle that runs through the rest of the product: the system makes the invisible visible, then puts the decision in your hands.

How moderation works

Moderation is automated. When a request is declined, the message tells you:

1. What happened and which policy applied
2. Where you can read the full policy
3. How to report the decline if you believe it was wrong

Every session has a Report button regardless of whether you are signed in.

We do not use vague or emotional language in moderation messages. The system is applying stated rules, and it describes itself that way.

Coop: human-in-the-loop moderation (coming later)

Coop is ROOST's human review platform. When it is active, human reviewers can triage flagged content, closing the loop between automated classifiers and human judgment for edge cases and appeals.

Coop human review platform is not live in the current alpha. It will be introduced in a future release once reviewer capacity is in place.

What we do not restrict

Questions about AI governance, AI sovereignty, government use of AI, and democratic accountability are core use cases for this product, not edge cases to be filtered. We do not treat political or governance content as inherently sensitive.

Your data

Conversations are stored only to operate the service — to show you your own history and keep the service running. Guest conversations are automatically deleted after 30 days, and the chat keeps no memory across sessions: nothing from one conversation is carried into another.

You can delete all your conversations, saved settings, and usage records at any time using the "Forget everything" button in the session bar. Deletion is confirmed plainly — no guilt prompts, no friction on the way out — and the system tells you exactly what it removed. Any safety reports you triggered are kept, detached from the conversation, for legal and safety compliance, and that count is disclosed to you.

Please do not enter sensitive personal information. We do not sell your data, and we do not use your conversations to train models without your consent.

What we collect, what we don't, and who processes data on our behalf is disclosed in plain language in our Privacy Policy.

Provenance and transparency

Every response shows the model name, organization, and compute infrastructure behind it. When the system uses web sources to answer a question, those sources are shown so you can verify the answer yourself.

This is an alpha. We say that plainly rather than presenting an incomplete product as finished. The gap map makes the full technical stack visible, including what is missing and what remains to be built.

Report something

Every session has a Report button. You do not need to be signed in to use it.