Autonomous work.
Clear boundaries.

Artie carries work forward with the access and instructions you give him.

You set the boundaries.

Choose the access, instructions, and ongoing work you give Artie.

Choose his access.

The permissions you grant in each app define what Artie can read, change or send.

Your instructions come first.

Outside content is reference material, not new permission. Action checks include AI judgment.

Artie
Competitor research
Working
Brief updated

Interactive illustration of pausing and resuming an example task. This control does not change a real account or task. Completed work stays complete.

Pause and resume work.

Pause work you manage, then resume when you’re ready. Completed actions stay completed; submitted actions may finish.

Your data, explained.

Doing your work and improving AI are different uses of information.

Used to do your work.

AI providers receive relevant context to carry out your requests. Processing and retention vary by provider and feature.

Clear limits on training.

We don’t authorize general model training on identifiable or confidential content. Eligible anonymous sharing defaults on where permitted. Admins can opt out.

Change access as work changes.

Disconnect accounts you no longer need, and request changes or deletion of saved information.

Questions about security.

What can Artie read and do?

Artie can use information you share, permitted workspace context and public sources. For connected tools, access depends on the account and permissions you authorize. Some permissions allow changes or sending messages as well as reading.

Custom connections can be personal or shared with your team. Sharing a connection and authorizing an action are separate controls. Give Artie a clear responsibility and review instructions, and limit permissions in the connected tool when an action must be technically restricted.

Saved logins use dedicated credential storage, with access scoped to authorized work. Authorized sign-in tools and browser sessions may hold the credentials needed to access your accounts. Use the connection flow rather than sharing passwords in chat.

When will Artie ask for approval?

Artie checks proposed actions against your request and standing instructions. Work you have delegated can continue without a separate approval at every step. If an action is held for approval, it waits for your decision.

Deciding what needs review includes AI judgment, which can make mistakes. “Ask me before sending” is an instruction, not a guarantee that every send will be held. An approval applies to the specific action presented for review.

Can a malicious email or webpage mislead Artie?

Artie is designed to treat outside emails, webpages, documents and tool responses as reference material, not new permission to act. Separate checks assess proposed actions against the work you authorized.

Those checks include AI judgment and do not make Artie immune to malicious instructions, often called prompt injection. Keep connected permissions appropriate to the work you want Artie to do.

Can I stop work that is already running?

Ask Artie to stop in Slack, or pause or cancel the relevant work in Artie. These controls apply to the work you are authorized to manage. Canceling a recurring task also stops its future scheduled runs.

Stopping does not undo completed actions. An action already submitted to another service may still finish. Check the affected tool for its result and contact our team if you need help reviewing what happened.

Who can see our information?

Workspace access depends on roles and sharing settings. Ordinary nonsensitive business facts explicitly shared in a direct message may become shared company knowledge. Raw direct messages and private preferences remain restricted to their appropriate scope.

Authorized staff and service providers may access relevant information for operations, support, security or legal needs. Our Privacy Policy explains these limits.

How do AI providers use our information?

AI providers and connected tools receive relevant context to carry out your work. Processing a request is different from training a general-purpose model. Providers may still process and retain information. Locations and retention vary by provider and feature.

We don’t authorize general model training on identifiable or confidential content. Eligible anonymous data may support training and improvement by default where permitted, with an admin opt-out. Covered Google Workspace data is excluded from general model training, including derived or anonymized forms. See our Privacy Policy for the full conditions.

What does the anonymous-data opt-out change?

Once effective, opting out stops new data being selected for optional improvement. Processing needed for your own work continues. Workspace admins can turn off “Share anonymized data” in Workspace preferences, or email hello@artie.ai.

An opt-out cannot necessarily remove data already irreversibly anonymized or undo lawful prior training. Our Privacy Policy sets out the conditions.

How do we disconnect access or delete our data?

Disconnecting withdraws usable access through that connection. It doesn’t automatically delete stored information, stop all scheduled work or cancel your plan. Revoking access directly with the provider may require a separate step.

To request deletion, contact hello@artie.ai. When the relevant service ends, customer data is returned or deleted under the Data Processing Agreement, ordinarily within 90 days, with limited backup and legal-retention exceptions. Shorter legal deadlines or source requirements take precedence.

Have a specific security requirement? Talk it through with our team.