Hacking group CyberCartel used AI agents built on Anthropic’s Claude, SpaceX’s Grok, and Moonshot’s Kimi to automate attacks on 400-plus Latin American banks, payment apps, and other financial targets, according to The Information. Grok agents researched targets, Claude agents planned each attack, and a Claude model inside Cursor ran the operation, which appears to have taken customer data and logins and gotten into bank accounts with little human effort. The group got past guardrails by telling the models they were in academic or company tests, and when Claude refused, the code passed the job to Kimi, then to ordinary malware. Anthropic said it blocked the group after Straiker reported it last week.
