The era of human-in-the-loop cyberattacks is officially over for some threat actors. Google’s cyber threat intelligence division detected a campaign where AI agents performed vulnerability discovery and exploitation entirely autonomously, marking the first known instance of a fully automated cyberattack without human intervention. According to a Politico report citing Google, Russian-speaking hackers deployed these agents to continuously compare software before and after security patches, identify flaws, and automatically develop exploits.

The Mechanics of Autonomous Exploitation

This isn't just AI assisting with reconnaissance or speeding up phishing email generation. The attackers created a series of AI agents that operated independently to bridge the gap between patching and exploitation. By analyzing code diffs, these agents pinpointed vulnerabilities in real-time and determined how to exploit them without waiting for a human operator to guide the attack or correct errors. This level of autonomy represents a significant shift from previous AI usage, which primarily accelerated stages like targeting and reconnaissance but still required skilled hackers to manage the operation.

Targeting European Infrastructure

The initial wave of these automated attacks targeted websites in France’s media, entertainment, and education sectors. While Google identified Russian-language technical documentation associated with the attackers, they stopped short of labeling the group as a state-backed entity. This distinction is crucial for infrastructure builders and security teams: the threat may originate from sophisticated criminal syndicates rather than nation-states, potentially altering how we approach attribution and response strategies. The broader context includes recent disruptions to refrigeration systems in supermarkets across nine EU countries, which cybersecurity specialists linked to possible Russian state-backed activity, though formal identification remains pending.

Key Takeaways

  • Google confirms the first fully automated AI cyberattack, removing the human from the loop for vulnerability discovery and exploitation.
  • AI agents analyzed software patches to identify and exploit vulnerabilities autonomously, targeting French media, entertainment, and education sectors.
  • Russian-speaking hackers were identified via technical documentation, but Google did not confirm state backing, distinguishing this from known groups like Midnight Blizzard.
  • Previous AI cyberattacks relied on human guidance for operations; this campaign demonstrated end-to-end autonomy in the attack lifecycle.

The Bottom Line

Security teams need to move beyond reactive patching. If AI can find and exploit vulnerabilities in the time it takes to deploy a patch, our current release cycles are a liability, not a feature.