The risk lies in what it can touch
An AI agent is not dangerous because it is intelligent, but because of what it can touch. Before connecting it to your email, CRM or client folder, ask the provider these seven questions, whoever they are. If any answer is vague, be wary.
The seven questions
- Can I choose which application each agent accesses? It is reasonable for the customer service agent to see email but not the accounts.
- Can I remove an access instantly? Without calling anyone or waiting for support.
- Are permissions enforced on the server? Hiding a button on screen is not a permission. What matters is that the system rejects the action even if the agent tries.
- Can I require approval before sensitive actions? Sending, publishing, paying or deleting should wait for a person to say yes.
- Is there a record of what it did, when and with which data? If something goes wrong, you need to be able to review it.
- Are my documents kept separate from other companies? Ask how each customer is isolated.
- Who is behind it and what do they do with my data? Company, country, privacy policy and whether your data is used to train models.
How we handle it at iabot
Each AI employee has permissions per application, enforced on the server, which you can connect or disconnect whenever you want. You mark which actions need your approval, and an activity log is kept.
Each company has its own workspace, with its own documents. On plans with a virtual computer, you can also watch the screen while the employee works.
Written by the iabot.io team. The questions apply to any provider, including us.