Google reports autonomous AI cyberattacks
According to Google's Threat Intelligence Group, attackers relied on AI agents to carry out the whole process of finding and exploiting vulnerabilities with no person involved, something Google calls a first.12 Google said the agents continuously compared code before and after patches, identified vulnerabilities and developed exploits automatically, and the attackers targeted French websites in media, entertainment and education.34 Politico says Russian-speaking hackers built the agent, and that it may be the first fully automated cyberattack.5
Politico notes that until now even advanced AI models could not run attacks entirely independently and often needed skilled hackers to guide them or fix mistakes.6 Politico reports that the discovery may represent a turning point for how AI gets used in cyberattacks. It adds that a prior autonomous hack, while it ran on its own, did not seem to be aimed with malicious intent at specific targets of the kind Google described.78
Google is quoted as saying that "the era of autonomous attacks has arrived" and that defending against machine-speed threats at human speed is not feasible.910 Commentator Zvi Mowshowitz expects attackers to hold the advantage because they can concentrate on weaker defenses such as unpatched systems.11 Google says its answer is Google AI Threat Defense, a system that pairs Gemini with other leading-edge models so that weaknesses can be located and patched before attackers take advantage of them.12
What each sentence rests on
fully automate vulnerability discovery and exploitation without a human in the loop for the first time.
— Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattackGoogle's Threat Intelligence Group has identified the first known case of fully automated cyberattacks using AI agents
— OECD AI Incidents Monitor, Google Uncovers First Fully Automated AI-Powered Cyberattacks - OECD.AIcontinuously compare pre- and post-patch code, identify vulnerabilities, and automatically develop exploits without human intervention,
— Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattackThe attackers targeted French websites in the media, entertainment and education sectors, Google said.
— Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattackRussian-speaking hackers built an AI agent that carried out what may be the first-ever fully-automated cyberattack, Google reported Monday.
— Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattack- Quote not reprinted: this page already repeats as much of this article as we allow. — Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattack
- Quote not reprinted: this page already repeats as much of this article as we allow. — Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattack
- Quote not reprinted: this page already repeats as much of this article as we allow. — Politico EU tech, Google says Russian speakers pulled off first-ever fully automated AI cyberattack
“the era of autonomous attacks has arrived.”
— Zvi Mowshowitz, AI #195: Opening Bell“Defending against machine-speed threats at human speed isn’t feasible,”
— Zvi Mowshowitz, AI #195: Opening Bellexpect attackers to have the advantage because they can focus their attacks where there is weaker defense, as in the unpatched systems.
— Zvi Mowshowitz, AI #195: Opening BellGoogle AI Threat Defense, which combines Gemini and other frontier models to find and fix flaws before attackers can exploit them.
— Zvi Mowshowitz, AI #195: Opening Bell