How Google used AI to stop a cyberattack before it even happened

As cyberattacks become more sophisticated, so do the tools used to stop them. Google has revealed that its artificial intelligence (AI) agent, Big Sleep, recently identified and blocked a major security threat before it could even be launched.

Developed by Google’s DeepMind and Project Zero teams, Big Sleep is an AI-powered agent designed to actively scan software for unknown vulnerabilities. Its job is to stay ahead of hackers, not by responding to attacks after the damage is done, but by spotting the warning signs and acting before any exploit can be used.

According to Google CEO Sundar Pichai, Big Sleep helped detect and neutralise what the company described as an “imminent exploit”. One of its biggest breakthroughs came when it uncovered a vulnerability in SQLite, a widely used database engine. The flaw could have been exploited on a large scale, but Big Sleep flagged it early, allowing Google to intervene before it became a threat.

This marks a major shift in cybersecurity strategy. Traditionally, defence teams have had to act reactively, patching systems after breaches occur. With tools like Big Sleep, the process becomes proactive, predicting attacks before they happen.

Google first introduced Big Sleep in 2023, and by November 2024, it had already recorded its first real-world security discovery. The company has since published a white paper outlining how privacy and responsible AI principles are built into the agent’s design. An issue tracker page also shares updates on vulnerabilities discovered and fixed by the AI.

In addition to Big Sleep, Google is developing other AI-powered defence tools, including FACADE (Fast and Accurate Contextual Anomaly Detection) and Timesketch, a collaborative digital forensics platform. Together, these tools aim to give defenders a real-time advantage in the fight against increasingly complex cyber threats.

As AI agents evolve, Google believes this could be a turning point, where machines not only support human teams but take the lead in preventing attacks altogether.