A Martinez man lost $315 in fraudulent gift card charges after a malicious add-on hijacked his Claude AI account, and his bank initially denied his fraud claim. The case exposes gaps in how AI platforms and financial institutions handle account takeovers that exploit third-party integrations.
How the account compromise unfolded
Jason Weitzman, a former information security director, bought an Anthropic membership to use Claude for building software at his tech company. Everything ran smoothly until Chase Bank flagged multiple gift purchases in euros - charges he never authorized. "Two had actually made it through to my bank, totaling around $315," Weitzman said. Additional charges appeared over the next two days.
Weitzman traced the problem to an add-on connected to Claude, referred to as a "skill," similar to a browser plug-in. "That basically told Claude to attempt to purchase different types of gift accounts on my stored information. So it was using the digital wallet that was on my computer for Claude to start to make these purchases," he said.
Anthropic's response and account lockout
Anthropic told 7 On Your Side the charges "appear to stem from a compromise of the user's device, not the Claude app itself." The company refunded Weitzman and restored his account access after the news team intervened. "In recent weeks we've also taken additional steps to further safeguard user payment information on Claude accounts," Anthropic said.
Before that resolution, Weitzman said he contacted Anthropic roughly two dozen times. After the company banned his account, he was locked out of live customer support and his appeal went nowhere. "You're frustrated, you're angry, you've had your money taken, nobody is helping you," he said.
The bank's denial and IP address dispute
Chase denied Weitzman's fraud claim, citing a Claude receipt as proof he authorized the charges. "I thought that was crazy. How could a receipt mean that I made those charges? It just means that a transaction occurred," Weitzman said.
Chase said its team "fully reviewed this claim including documentation that was made available to us at the time. We asked Mr. Weitzman to provide us with evidence of hacking or malware. He declined to provide any additional documentation." The bank added that it "determined that the transaction was made from the customer's known IP address." Weitzman disputes that Chase ever requested hacking evidence and pointed out that IP addresses are commonly spoofed during account takeovers.
Why this matters for customer support teams
This case highlights three friction points support professionals should address. First, when a platform bans a user's account during a fraud investigation, it cuts off the primary channel for resolution - leaving customers with no way to reach a human. Second, banks relying on receipts and IP addresses as proof of authorization creates a gap that skilled attackers exploit. Third, third-party integrations like Claude's "skills" introduce attack surfaces that neither the AI provider nor the user may fully control.
For teams building or managing AI-powered support workflows, the takeaway is concrete: fraud detection systems must account for compromised integrations, not just direct login attacks. Customer-facing teams also need clear escalation paths that survive an account ban. Professionals looking to build these skills can explore AI for Customer Support Courses that cover automation risks and security-aware support design.
Your membership also unlocks:
