Who’s responsible when an AI assistant makes an unwanted purchase?

Banks and merchants typically treat transactions initiated by AI agents as if they were directly authorized by the account holder. As Aaron Painter, CEO of Nametag and author of Deepfaked, explains, “It’s the new refund trap. If you authorize AI to act, the banks view it as your responsibility.” This stems from a lag in policy updates failing to keep pace with rapidly evolving technology, prioritizing convenience over accountability.

The core issue lies in a lack of robust verification processes. Retailers often rely on initial login credentials without requiring repeated authentication, leaving accounts vulnerable to unauthorized access. “When you log into a website, you provide personal information, but then you don’t have to re-verify anything,” Painter notes. “Someone could literally access your account, and the retailer can’t help but think it’s you.”

This vulnerability is amplified by the increasing popularity of AI assistants like Amazon’s Alexa for Shopping. While convenient, these tools are susceptible to hacking or device cloning, potentially leading to unauthorized purchases. Painter emphasizes that the trustworthiness of AI is directly tied to the security measures in place.

To mitigate these risks, Painter recommends several protective measures: granting AI assistants read-only access, reviewing bank terms regarding third-party providers, establishing dedicated checking accounts with limited balances, and utilizing credit cards with purchase protection, such as those offered by American Express. Real-time account updates and prompt action – including contacting banks, disputing charges in writing, and revoking agent access – are also crucial.

He also advises caution regarding urgent requests, suggesting a pause and verification through secure channels. Sharing experiences with others is vital, as collective awareness can help combat increasingly sophisticated scams. While legislative protections are anticipated, proactive measures and shared knowledge remain the best defense against erroneous AI purchases.