AI use policy · Cloverr

Cloverr includes AI features: agents ("Bees") that answer questions, draft emails and documents, summarise calls, create tasks and run automations. This policy explains how they work and the rules for using them. It sits alongside the Terms of use, the Privacy policy and the Acceptable use policy.

How the AI works

Cloverr's agents use large language models provided through specialist AI providers. Your organisation can also connect its own AI provider key, in which case that provider processes the requests under your organisation's agreement with them.
When an agent works on something, the request and the information it needs (for example the email it's answering or the documents you pointed it at) are sent to the model to produce an answer. Providers we use are not permitted to train their models on this data.
Your workspace content is never used to train AI models, by us or for us.

What an agent can see

Each agent only reads what its access settings allow (for example "read mail" or "read docs"), and only within your organisation.
Private mail and personal calendars stay private to their owner. Agents reading the web is a separate permission your organisation's owners control.
Every agent action is logged on its Trail, so you can see what it read and what it did.

AI makes mistakes

AI output can be wrong, incomplete, out of date or inappropriate, even when it sounds confident. Check anything important before you rely on it, send it or publish it.
You're responsible for what you choose to send or publish, including text an agent wrote for you.
Agents are told not to invent facts about your organisation and to say when they don't know. That reduces mistakes but can't remove them.

Automatic checks and decisions

Outgoing email: before an email leaves your organisation, an AI reviewer checks it. Emails that look risky (for example leaked passwords, phishing or abuse) are held for a person, and replies that don't answer what was asked are sent back to be rewritten.
Safety screening: text that strongly suggests scams or abuse can pause an account while a person reviews it.
Untrusted content: emails, web pages and form answers from outside are treated as information, never as instructions, to guard against attempts to manipulate the AI. Messages that try are held for a person.
You can ask a person to look again at any automated decision by emailing kyle@imkylejk.me.

Rules for using the AI

Don't use the AI to create anything the Acceptable use policy forbids, including scams, harassment, malware or illegal content.
Don't try to make an agent reach data it hasn't been given, reveal its instructions, or get round its checks.
Tell people when they're talking to or receiving messages from an AI where that matters. Cloverr's agents say they're AI if asked.
Don't rely on AI output for medical, legal, financial or other professional advice.

Limits

To keep the service fair and fast, there are limits on how much AI each person and organisation can use in a short time. Very large bursts of outside email may be left for a person to answer rather than answered by AI.

Changes and questions

We'll update this page as the AI features change. Questions: kyle@imkylejk.me.