Google’s AI tool Big Sleep has discovered a critical zero-day vulnerability in SQLite and has successfully blocked its active exploitation

Google’s cutting-edge AI-driven security tool, Big Sleep, has reached a major milestone by identifying and halting the exploitation of a critical SQLite 0-day vulnerability, making it the first instance where an AI agent has actively neutralized a live cyber threat.

The uncovered flaw, CVE-2025-6965, was previously unknown to the public and exclusively leveraged by threat actors—highlighting the game-changing role of AI in modern cybersecurity defense.

Key Takeaways:

1. Google’s Big Sleep AI detected and blocked a critical SQLite 0-day vulnerability (CVE-2025-6965) before it could be exploited.
2. It marks the first instance of an AI agent proactively stopping live cyber threats in real time.
3. Leveraged threat intelligence and predictive analysis to anticipate attacks and trigger preemptive mitigation.
4. Highlights a major shift in cybersecurity—from reactive response to proactive AI-driven defense.

Big Sleep AI Prevents SQLite 0-Day Exploitation

Big Sleep, an AI agent jointly developed by Google DeepMind and Google Project Zero, successfully uncovered the critical SQLite vulnerability CVE-2025-6965 using advanced threat intelligence analysis.

This severe security flaw posed a major threat, as it was previously undisclosed and known only to malicious actors actively preparing to exploit it.

Affecting SQLite, one of the most widely deployed database engines across numerous applications and systems, the vulnerability had far-reaching implications.

Big Sleep identified the flaw by analyzing data from Google Threat Intelligence, leveraging deep insights into evolving attack patterns. Using sophisticated pattern recognition and vulnerability assessment algorithms, the AI agent accurately predicted that the SQLite exploit was on the verge of active use by attackers, enabling timely mitigation.

This proactive discovery enabled Google’s security team to deploy immediate defensive measures and collaborate with SQLite developers to patch the vulnerability before any real-world exploitation could take place.

The milestone marks a paradigm shift in cybersecurity, as Big Sleep has not only met but surpassed expectations, significantly accelerating AI-driven vulnerability research since its launch.

Unlike traditional scanners, Big Sleep uses predictive analysis and real-time threat assessment to uncover zero-day vulnerabilities before they are weaponized—offering a major leap forward in threat prevention.

This achievement builds on the AI system’s growing track record, following its first real-world vulnerability discovery in November 2024, which showcased the tremendous potential of AI to preempt security breaches.

By processing massive volumes of security data, Big Sleep has proven instrumental in safeguarding both Google’s ecosystem and widely-used open-source software from emerging threats.

According to Google’s report, the initiative reflects a strong commitment to responsible AI deployment, emphasizing secure-by-design principles, human oversight, and transparency in automated operations.

This breakthrough sets a new benchmark for proactive cyber defense, with the potential to reshape how organizations counter advanced and evolving cyber threats.