A hacking group known as CyberCartel used a network of AI agents to attack more than 400 banks and payments companies in Latin America, according to research by the security startup Straiker shared with The Information, per PYMNTS. The agents ran on models including Anthropic's Claude and Grok. They helped steal passwords and customer data and got into customer bank accounts on their own, with little human effort.
Straiker has not publicly disclosed its full report, so the 400-target count and its method can't yet be checked against the source. The Information also named models from China's Moonshot AI. Anthropic said it opened an investigation after the findings surfaced and blocked the group's use of Claude. No response from Grok's owner was reported.
CyberCartel is not new. Darktrace has tracked it targeting users of banking and government services in Latin America, especially in Mexico and Chile, and IBM describes it renting existing malware instead of writing its own. What changed is the labor. Agents now do the slow parts of the job, such as testing stolen credentials and working through account screens, at machine speed. South Korea's bank breaches this month followed the same pattern with a different tool, as we covered.
Bottom Line
Assume account-takeover attempts against your customers now run around the clock and adapt between tries. Rate-limit login and recovery flows per device and per account, and put step-up checks on new payees, because that is where an agent turns a stolen password into money.