Can AI hack you — or even attack you?

For decades, a serious cyberattack needed a skilled human team, weeks of patient work and a lot of money. That barrier is falling fast. AI models can now read code, find its weak spots, write working attacks and run them — at a speed no human team can match. For a business like ours, which lives on connected devices, online orders and cross-border payments, this changes the risk picture completely.

Can AI hack you? It already has

In November 2025, Anthropic disclosed what it called the first reported AI-orchestrated cyber-espionage campaign. A group it assessed as Chinese state-sponsored, labelled GTG-1002, turned the Claude Code agent into a semi-autonomous attack tool aimed at around 30 organisations: tech firms, banks, chemical manufacturers and government agencies (Anthropic report).

The numbers are what matter. The AI carried out roughly 80–90% of the tactical work on its own — scanning networks, writing exploit code, harvesting passwords and sorting stolen data. Humans only picked targets and approved key steps. The attackers tricked the model by posing as a legitimate security firm and splitting the job into harmless-looking pieces. The campaign succeeded against a small number of targets before Anthropic detected it, banned the accounts and notified victims and authorities (Cyber Magazine, IAPS analysis).

Some security experts questioned how “autonomous” the operation really was, pointing to limited public technical evidence (Giskard). But the direction is clear: the human hacker is becoming a supervisor, and the AI does the legwork.