Concerns AI email agents could be fooled by scammers
Updated: 5 days ago
A small business owner has been expressing concerns that AI email agents designed to help with the influx of emails may not be a catch-all solution.

In a blog post on BitDefender, small business owner Christina Popov concedes that inbox fatigue can be a real problem for small businesses and using AI can help - however, she also warns that giving AI access to your business inbox could lead to security and privacy risks.
Popov says scammers are starting to create emails which are designed to target AI email agents directly in a variation of the phishing email.
Described as prompt injection, the email contains instructions that an AI agent will blindly follow, such as forwarding confidential information to another address.
Popov writes: "If the agent treats those instructions as commands rather than untrusted email content, it could take actions you never authorized. The risk is greater when it can send emails, access files, or interact with other business systems."
She also claims AI could miss what a human would immediately recognise as fake invoices or supplier information, as well as fraudulent payment requests.
An example Popov offers could be when a scammer impersonates a supplier and asks for bank details to be updated to a different account, which the AI could follow without verification if it has access to supplier records.
In her blog, Popov does offer some advice for small business who are, or are considering using an AI email agent:
Start with AI assistance, not full automation. Use AI to summarize emails, organize messages, and prepare drafts. Once you're comfortable with how it works, consider automating a few low-risk tasks.
Limit what the agent can access. If it only needs to read messages and prepare drafts, don't give it permission to send or delete emails. You could also use a separate customer service inbox rather than connecting your main business account.
Keep a human involved in sensitive decisions. Payment requests, changes to bank details, contracts, refunds, and requests for confidential information should always require your review and, where appropriate, independent verification.
Choose your AI provider carefully. Check its privacy policy, security practices, data retention rules, and whether you can revoke access or delete stored information.
Review what the agent is doing. Check automated replies, activity logs, and unusual actions regularly. Make sure you can pause the agent or disconnect it quickly if something goes wrong.
Protect your email account and business devices. Use strong, unique passwords, enable multifactor authentication, and keep your devices protected against phishing, malware, and other online threats.






Comments