Security and data handling
An agent connected with an API key can read only the project the key belongs to, and can change only the status of content actions. We check every request against the key's scopes and its creator's current role, and we record every capability call.
What an agent can and cannot reach
One project. The project always comes from the key. An agent cannot read or change another project or another organization, and it cannot request aggregated data from other Ansehn customers.
Read-only by default. All capabilities read data, except updating the status of a content action, which needs the separate Content Actions Write scope.
No other changes. Keys cannot edit personas, prompts, monitors, competitors or buying-simulation schedules, cannot start simulations or spend credits, and cannot send Slack or Teams messages. Those changes happen only in Copilot inside Ansehn, after a person confirms them.
Never more than its creator. A key acts with its creator's current permissions on the project. When the creator loses access, the key stops working.
Assistants connected by signing in
An assistant you connect by signing in (Claude, ChatGPT) follows the same rules as a key: one project, the permissions of the person who connected it, the same capabilities and the same audit trail.
The project comes from your choice on the Ansehn sign-in page, never from the assistant.
Short-lived access. The assistant receives an access token that expires after 15 minutes and is renewed automatically. We check on every request that the connection is still active and that you still have access to the project.
Disconnect any time in Project Settings > Connected apps. The assistant loses access on its next request. See Connect Claude or ChatGPT to Ansehn.
No passwords are shared. You sign in on our own page; the assistant never sees your password or an API key.
Text from outside your account
Results can contain text we did not write: AI answers, simulation transcripts, citation snippets and page titles. We wrap that text in external_content tags. Instruct your agent to treat it as data and never follow instructions found inside it. The text cannot change which project the key reaches or grant it more permissions.
What we store
API keys are stored as a one-way hash. We cannot show or recover a key after it is created.
Every capability call is recorded: the key or connected app, the person it acts for, the capability, the result, and a fingerprint of the input. We never store the key, tokens or the raw input.
Discovery requests (listing capabilities, the OpenAPI document, and listing MCP tools) and requests with an invalid key are not recorded as capability calls.
Good practice
Prefer signing in for Claude and ChatGPT over pasting an API key.
Create one key per agent or integration, with only the scopes it needs.
Keep keys in a secret store or environment variable, never in source control or shared documents.
Set an expiration date for keys used in experiments.
Revoke a key as soon as you stop using it or suspect it was exposed. Revoking takes effect immediately.
If you have questions about security, contact us at [email protected].