Google: AI agents can now carry out cyberattacks in under six hours

Google has published a new threat report showing a shift in cyberattacks: from AI merely assisting attackers to AI agents now carrying out large parts of the attack themselves.

In one specific case, attackers broke into a cloud service and then let AI agents plan, build and carry out a large theft of login credentials, all in under six hours, without much human involvement along the way.

The report also points to broader trends: AI is now used for everything from tracking down vulnerabilities and writing custom malware to automating social engineering and spreading disinformation at scale. At the same time, attackers are going after the AI tools themselves (code, models and login credentials) to gain access without paying for it.

For ordinary users, nothing changes tomorrow. But for businesses, the point is simple: the time you previously had to detect and stop an attack has narrowed significantly. Security teams that still think in days should start thinking in hours.

Takeaway: If your company handles cloud infrastructure, customer data or developer logins, it is now relevant to ask your security vendor whether its solution can respond automatically, not just alert. Private users can relax, but as always should keep track of passwords and two-factor login.