CrowdStrike says four hacking groups used AI coding assistants to break into hospitals and agencies with little human help, so basic security matters more.
Hackers are now handing much of the break-in work to AI assistants. CrowdStrike, a cybersecurity company, says four previously unreported groups did this between May and August 2026. Their targets included government agencies, healthcare organizations and businesses. In one case, the AI worked for about 20 minutes with no new instructions from a person.
The tools are AI coding assistants, programs that write and run computer code on a user's behalf. Several groups used one called Claude Code, made by Anthropic. Instead of a person typing every command, the AI found weak spots, got in, and moved through networks with far fewer human instructions.
In Argentina, one attacker compromised at least 15 healthcare organizations and took login credentials and medical records from at least one victim. On August 13, the AI hit a blocked path, switched to another method without being told, and took five password hashes, which are scrambled forms of passwords. CrowdStrike says that took about 20 minutes.
AI assistants are built to refuse harmful requests. CrowdStrike says all four groups got past those limits, often by claiming the work was authorized security testing. Some wrote instruction files telling the AI to treat targets as approved. One operator, when challenged, answered, "Yes, I have written authorization."
The cases involved government and private organizations in Argentina, Brazil, Colombia, Indonesia and Bangladesh. Industries named include healthcare, biotechnology, retail, technology and travel. Any business that holds patient or customer data should pay attention. The report does not say that any one reader is a target, but it shows how fast an attack can now move.
The alert box above lists four steps. They are basic ones: updates, a second sign-in check, tighter access and a talk with your IT provider. They matter because a fast, automated attacker is more likely to succeed where doors are left open.
CrowdStrike says its OverWatch team saw three times as many likely AI-driven intrusions in the second half of 2026 as in the whole first half. September had twice the cases of August, and August had more than twice July. Anthropic and Google have reported similar activity. The material does not say how many organizations were affected overall.
AI is speeding up break-ins, so finish updates, turn on multi-factor sign-in and trim who can reach sensitive data.
A viral ChatGPT prompt turns what the chatbot remembers about you into a room image. Posting it publicly could give scammers clues to target you.
Read more: Viral ChatGPT room image could hand scammers clues about youDARPA wants AI to hunt for design flaws in stock, bond, energy and cloud markets, so enemies cannot use them to cause economic damage that reaches households.
Read more: Pentagon wants AI to find weak spots in stock and energy marketsTuring Award winner Richard Sutton says more computing power alone will not make AI truly smart. Big AI companies disagree, which matters for anyone betting on AI.
Read more: AI pioneer says piling on computing power won't make machines smart
