A Martinez tech entrepreneur and former information security director says he received several fraudulent charges while using a Claude add-on, and while Anthropic denied responsibility, it agreed to reimburse him after media inquiries.
Jason Weitzman told ABC 7 on Your Side’s Stephanie Sierra that he discovered hundreds of euros in fraudulent gift card purchases and traced the activity to a third-party "skill," or add-on, connected to Claude, which he believes accessed payment information stored on his computer.
Weitzman said his attempts to resolve the issue quickly unraveled on both sides. He told the outlet he was locked out of his Anthropic account after the company flagged and ultimately banned it during his support requests, cutting off access to live customer assistance.
At the same time, he said Chase Bank denied his fraud claim, pointing to Claude-related transaction records and activity tied to his known IP address as evidence the charges were authorized. Weitzman disputed that conclusion, telling ABC 7 he was never asked for additional proof of a compromise and that standard fraud indicators can be misleading in account takeover cases.
Anthropic told the outlet it found no evidence that Claude itself had been breached, instead attributing the charges to a compromise of Weitzman's device.
After 7 On Your Side intervened, Anthropic restored his access and refunded the disputed charges.
“The user in question has been refunded and their access to Claude has been restored,” said the company. “In recent weeks we've also taken additional steps to further safeguard user payment information on Claude accounts.”
In response to the incident, Weitzman concluded that while Claude is a great tool, “the infrastructure around that tool, not so great.”
The incident comes as Anthropic continues to tighten security around its AI systems amid growing concerns over their cybersecurity implications. The company reportedly says it has added additional protections designed to detect harmful requests and block malicious activity, including ransomware development and large-scale data theft.
SFist reported in April that leaked internal documents showed the company believed a forthcoming Claude model could pose unprecedented cybersecurity risks.
**Related: **Anthropic's New Model, Mythos, Is So Dangerous It Isn't Being Released to the Public
Image: Michael M. Santiago/Getty Images