Google said Russian-speaking hackers used AI agents to carry out what may be the first fully automated cyberattack, with no human involvement in finding software vulnerabilities and exploiting them.
Google’s cyber threat intelligence division said it had observed hackers using AI to “fully automate vulnerability discovery and exploitation without a human in the loop for the first time.” The company said the attackers used Russian in technical documentation.
According to Google, the hackers set up AI agents to “continuously compare pre- and post-patch code, identify vulnerabilities, and automatically develop exploits without human intervention.” The agents independently found software flaws and ways to exploit them.
Google said the attackers targeted French websites in the media, entertainment and education sectors.
The report comes amid wider concerns about AI in cyberattacks, following the release of Anthropic’s specialized model Mythos in April. The article said advanced AI models had previously been unable to conduct cyberattacks entirely independently, often requiring skilled hackers to guide them or correct mistakes.
The article also cited an autonomous hack by an OpenAI agent on the Hugging Face platform this summer, noting it did not appear to be maliciously aimed at specific targets like the attack Google reported. It also mentioned recently discovered major flaws in PHP, which the article said underpins 70 percent of the web, and hacks on supermarket fridges partially blamed on Iran.
