Is your AI agent safe? What to check before connecting your ad accounts

Connecting an AI agent to your ad accounts and payment processor is a significant trust decision. If you get it wrong, you're handing over OAuth access to systems that control real money and real customer data. Here are the questions to ask before you connect anything.

1. Does it use OAuth or does it ask for credentials?

Any legitimate AI agent uses OAuth — the platform-native auth flow that gives the agent a scoped access token. If a tool asks you to paste your Meta Ads username and password, your API key, or your Stripe secret key into a chat interface, close the tab.

OAuth means the credentials never leave Meta's servers. The agent gets a token that can be revoked at any time. It's the secure path.

2. What scopes does it request?

OAuth tokens are scoped. A read-only scope can only read data. A write scope can make changes. Before you authorize any agent, check what permissions it's requesting. A reporting tool that needs 'manage campaigns' access should be a red flag — why does it need to change your campaigns to show you data?

• →Read-only: can see data, can't change anything
• →Read+write: can both view and modify
• →Admin: potentially can add/remove users, change billing settings

3. Does it require approval before taking action?

This is non-negotiable. Any AI agent that can take action on your accounts — pause campaigns, change budgets, update CRM records — must show you the proposed action and wait for your explicit approval before executing. No exceptions.

If a tool offers 'fully automated' campaign management with no approval step, it can burn your budget without a human seeing it first.

4. Does it store your data or train on it?

Ask directly: does your query data get stored? Is it used to train models? Is it shared across customers? You want a vendor that can answer all three with an unambiguous 'no' — in writing.

5. Can you revoke access immediately?

You should be able to disconnect an integration or remove the agent from Slack entirely and have all access revoked within minutes. If a vendor can't explain their revocation process, that's a concern.

6. Where is data processed?

For EU/UK businesses, GDPR compliance requires knowing where data is processed. Ask about their sub-processors and whether they have a DPA available.

7. What's their incident response process?

If their systems are compromised, how quickly will they notify you? 24 hours is a reasonable bar for a security incident affecting your connected accounts.

8. What happens when you cancel?

When you cancel the service, what happens to your OAuth tokens? Are they deleted? What about your conversation history and any saved templates? Get clarity before you sign up.