A Secure Summer: Equipping Cyber Defenders Through AI
Sure! Here’s the rewritten content with promotional references adjusted accordingly:
- Giving Defenders an Edge with Agentic Capabilities
Last year, we introduced Big Sleep, an AI agent developed by Google DeepMind and Google Project Zero, that actively searches for and identifies unknown security vulnerabilities in software. By November 2024, Big Sleep successfully uncovered its first real-world security vulnerability, demonstrating the significant potential of AI to address security gaps before they affect users.
Since then, Big Sleep has continued to find multiple real-world vulnerabilities, exceeding our expectations and accelerating AI-powered vulnerability research. Most recently, leveraging information from Google Threat Intelligence, the Big Sleep agent discovered an SQLite vulnerability (CVE-2025-6965) — a critical security flaw previously known only to threat actors, putting it at risk of exploitation. By combining threat intelligence with Big Sleep, we were able to predict that this vulnerability would be imminently targeted and proactively mitigate the threat.
We believe this marks the first instance of an AI agent being employed to directly thwart attempts to exploit a vulnerability in real-time. These AI advancements not only enhance the security of Google’s products, but Big Sleep is also being applied to bolster the security of widely used open-source projects — a significant achievement that promotes faster and more effective security across the internet at large. These cybersecurity agents are a game changer, enabling security teams to concentrate on complex threats, thereby vastly increasing their impact and reach.
Of course, this work must be conducted safely and responsibly. In our latest white paper, we outline our approach to building AI agents in a manner that prioritizes privacy, mitigates the risks of unintended actions, and ensures the agents operate under human oversight and transparency. When deployed according to secure-by-design principles, these agents provide defenders with an unparalleled advantage.
We will continue to share our insights and findings regarding agentic AI through our established disclosure process. Keep informed about all publicly disclosed vulnerabilities from Big Sleep via our issue tracker page.
Feel free to let me know if you need any further adjustments!
