AI-Driven Hacking Breaches 400+ Banks, Downs UK Government Sites and Slips Into US Bank Accounts

Three separate security stories landed inside the same 48 hours this week, and together they describe a single shift: AI-directed attacks have moved from theory into live breaches of banks, government infrastructure and ordinary customer accounts.

A HACKING GROUP LET AI MODELS RUN THE ATTACK ON 400+ BANKS

Security firm Straiker disclosed that a criminal group called CyberCartel used a system it built, nicknamed CARLIY0S, to coordinate attacks against more than 400 banks and payment providers across Latin America. The group assigned AI models from Anthropic, xAI and Moonshot AI distinct roles inside the operation, using them to plan targets, develop attacks and coordinate the campaign rather than simply asking a chatbot for help with one task at a time. Straiker's research, shared with The Information, documented over 200 exploited vulnerabilities and a harvested cache that included payment settlement credentials and hundreds of bank account numbers. Anthropic said it opened an investigation once the findings surfaced and cut the group's access to Claude.

A PHP BUG TOOK DOWN DOZENS OF GOVERNMENT WEBSITES

The same week, the PHP project rushed out emergency patches for two critical flaws, nicknamed PHP2Shell, that let an attacker run code on a web server without logging in. Both carried a severity score of 9.2, details leaked publicly before the fix shipped, and Amazon's threat intelligence team later traced exploitation attempts back to four days before the public disclosure. Because PHP runs underneath close to 70 percent of the web, the fallout reached governments directly. UK tech minister Baroness Lloyd of Effra told the House of Lords that 32 government services were unavailable and four more partially disrupted as departments pulled sites offline as a precaution, with the DVLA and police services among those affected. Germany, France, Belgium, Finland, the Netherlands and Australia's own cybersecurity agency took similar precautionary steps with their own government sites.

US BANKS FOUND AI AGENTS ALREADY INSIDE CUSTOMER ACCOUNTS

Axios then reported that major US banks, including Morgan Stanley, JPMorgan, Bank of America and Wells Fargo, are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents attempting actions that appear independent of what the customer actually asked for. Bank executives told Axios they are struggling to tell a human login from an AI agent, and struggling further to tell an agent the customer authorized from one acting on its own. Several banks are now preparing for what they describe internally as worst-case and "day after" scenarios, covering how to rapidly lock down or restore accounts if an agent causes real damage. Morgan Stanley has moved to block third-party agents outright, while JPMorgan, Bank of America and Citi are restricting their access.

BITCOIN SAW A VERSION OF THIS THREE MONTHS EARLIER

On the show, the hosts connected this week's cascade back to a run of Bitcoin-specific incidents that started in late July, when a five-year-old firmware flaw in Coinkite's Coldcard hardware wallet let attackers drain roughly 1,816 BTC from more than 5,200 addresses by exploiting weak randomness in how the device generated seed phrases. That incident, along with a string of Lightning Network project vulnerabilities in the months after, pushed parts of the Bitcoin development community into an informal security sweep the hosts referred to on air as a red team effort. The hosts argued Bitcoin absorbed this kind of AI-assisted probing first precisely because it is a live, bearer-asset target with no central operator to patch things quietly, and that the same scanning pressure was always going to reach banks, governments and the rest of the internet next. This week, it did.

This story comes from the Simply Bitcoin Live show. Watch the full episode.