AI agents and your business data: what actually happens to it

Handing account access to something that acts on its own is a real decision. It deserves specific answers rather than reassurance.

What actually leaves your building

For an agent to write your post it must see enough context to write it. That content goes to a model provider for processing. That is unavoidable in any AI system and honest vendors say so rather than implying the data never moves.

The training question

Ask directly: is my data used to train models? Major providers contractually commit not to train on data submitted through their business APIs — but that commitment has to be passed through by whoever is in the middle. A vendor who cannot answer this crisply has not read their own contracts.

Credentials are the real exposure

The content is one risk; access is the bigger one. What matters is how tokens are stored, whether an agent can reach a tool it has no reason to touch, and whether you can revoke without asking anyone.

• Never give a vendor a password. Access should go through each platform's own permission screen.
• Ask whether credentials are scoped per agent or shared across all of them.
• Confirm you can revoke at the source, instantly, without a support ticket.
• Ask whether secrets appear in prompts by default.

Our answers to all of the above are on the security page and in the privacy policy.

Keep reading

• Every agent platform we get compared to, side by side