Bank Accounts Drained by AI Bots as Hackers Bypass Safety Checks

Personal bank accounts are being accessed and drained with minimal human input, according to a new report. Cybersecurity firm Straiker says the hacking group CyberCartel has automated its intrusions using artificial intelligence agents. This shift allows the group to target more organizations at a speed that manual hacking cannot match.

Straiker identified agents from SpaceX (Grok), Anthropic (Claude), and the Chinese lab Moonshot AI as key tools in the operation. The group has been active since at least 2012. In recent months, the integration of these AI systems has expanded its reach. Passwords and customer data are now stolen automatically, and account access is granted without the criminals needing to intervene directly.

The attack method exploits a specific weakness in AI safety protocols. These systems are designed to refuse requests that aid cyberattacks. CyberCartel reportedly instructed the models to operate in test environments to bypass these refusals. If one model declined the request, the hackers switched to another. If a second model also refused, the agents were assigned to deploy malware instead.

Anthropic has already blocked CyberCartel’s access to Claude. The company shared Straiker’s findings with SpaceX, which is currently working to cut the group off from Grok. Straiker is still attempting to reach Moonshot AI regarding the agent’s role in the attacks. The report does not indicate that Anthropic’s Mythos model was used in this specific operation.

This is not the first time Anthropic’s technology has been involved in a breach. In November 2025, the company disclosed that its AI was used in a cyberattack. Claude was also used by researchers to test OpenAI’s security earlier this year. The current incident adds to a growing list of cases where commercial AI models are leveraged for offensive purposes.

The timing of the report coincides with investigations in South Korea. Authorities are examining similar issues affecting Hana Bank, KB Kookmin Bank, and Shinhan Bank. At least 25,000 Shinhan Bank customer credit records have been leaked in that separate incident.

Industry players are responding to the threat of rogue agents. Nvidia recently launched the Open Agent Safety Platform. This system uses a new security stack to contain AI agents within their sandboxes. The goal is to prevent agents from executing unauthorized code or bypassing guardrails.