Google’s Agentic Vulnerability Discovery Harness Strengthens AI Security, Identifying 100+ Critical Vulnerabilities in 10 Months

The emergence of adversarial AI has significantly heightened the cybersecurity threat landscape, particularly in terms of data theft and extortion. As proprietary source code becomes increasingly susceptible to attacks, defenders often find themselves racing against time to patch vulnerabilities while adversaries exploit AI tools that operate at unprecedented speeds. In light of these challenges, Google has unveiled the Agentic Vulnerability Discovery Harness (AVDH), a framework aimed at enhancing AI security by optimizing the vulnerability discovery process.

AVDH integrates structured analysis, rigorous validation steps, and specialized human expertise to shift the advantage toward defenders. This innovative framework combines AI models with a human-driven orchestration layer, allowing security teams to identify and remediate vulnerabilities before they can be exploited. The internal architecture of AVDH is now being shared publicly, enabling organizations to adopt similar strategies to strengthen their defenses.

Real-World Impact of AVDH

Since its implementation, AVDH has proven effective in real-world applications. Within just ten months, the harness has played a crucial role in identifying over 100 critical vulnerabilities during an incident response investigation involving compromised corporate repositories. This rapid identification process was accomplished in only two days, a significant improvement over traditional manual reviews.

The framework has shown particular effectiveness in analyzing large codebases. Mandiant has utilized AVDH to examine tens of millions of lines of code and execute thousands of pipelines, resulting in tens of thousands of findings. This swift analysis has led to the identification of numerous assignable flaws in widely used web extensions and open-source projects, culminating in the assignment of 12 Common Vulnerabilities and Exposures (CVEs), including CVE-2026-13242 and CVE-2026-55803, with additional vulnerabilities currently under active disclosure.

In addition to its speed and accuracy, AVDH has served as a force multiplier during targeted adversary simulation engagements. For instance, the harness swiftly identified a remote code execution (RCE) vulnerability in a client’s web application source code, facilitating initial access for further testing. This capability highlights AVDH’s value in navigating complex exploit chains and overcoming advanced defenses.

Architecting the AVDH Pipeline

The architecture of AVDH is grounded in the principles of leveraging large language models (LLMs) for cybersecurity applications. By addressing the unpredictability associated with LLMs, AVDH enhances their effectiveness in code analysis. The programmatic infrastructure orchestrates agents in a deterministic manner, ensuring that each phase of the analysis pipeline is completed before progressing to the next. This structured approach produces a prioritized, risk-rated list of findings that are ready for human review.

To achieve this, AVDH utilizes the Google Agent Development Kit (ADK), which facilitates common agent orchestration patterns and allows for custom integrations. This framework aligns with the capabilities offered by Google Antigravity, providing a centralized workspace for managing agentic workflows.

Furthermore, the effectiveness of AVDH is bolstered by the contextual data it employs, including asset inventories, software bills of materials (SBOMs), architecture documentation, and threat intelligence. This comprehensive environmental input enables agents to dynamically select relevant skills and vulnerability patterns for in-depth analysis, ultimately enhancing the accuracy and relevance of the findings.

Implications for Cybersecurity

As organizations continue to confront the challenges posed by adversarial AI, frameworks like AVDH represent a significant advancement in the ongoing struggle for cybersecurity. By merging human expertise with sophisticated AI tools, defenders can better anticipate and mitigate potential threats, contributing to a more secure digital environment.

For further insights into this innovative approach, additional information can be found from Google’s research team here.