Stop supplier bank-detail fraud in AI-automated finance
“Our bank details changed” is the oldest invoice fraud there is. When an AI agent processes supplier emails, the attacker no longer has to fool a person. They have to fool the agent.
Gmail
NetSuiteThe workflow
A supplier email announces a new IBAN. The agent finds the vendor in NetSuite and prepares the change for review instead of applying it.
The attack hidden in it
A hidden comment tells the agent to update the vendor's bank account now because it is “pre-approved” and to skip the approval step.
How the guardrails handle it
- Screening withholds the hidden instruction; the agent sees only the visible request.
- Bank-detail changes are writes, so they always wait for a person, whatever the email claims.
- The approval shows the exact vendor update, and approving sends only those bytes, once.
Same workflow, every company size
The agent and the guardrails stay the same; only the systems behind them change.
Sage Intacct
Microsoft Dynamics 365 Business Central
Exact Online
Xero
Moneybird
QuickBooksQuestions
Can an AI agent change vendor bank details?
It can prepare the change, but a person must approve it. Approvals can't be skipped by anything the agent reads.
How do I detect bank-detail change fraud with AI?
Treat every change request as a write that needs approval, screen the email for hidden instructions, and verify the new details out of band.
Which ERPs does this work with?
NetSuite, Sage Intacct, Microsoft Dynamics 365, QuickBooks, Xero, Exact and the other accounting systems Apideck connects.