Anthropic and Nozomi Unveil AI Vulnerability Research for Enhanced OT Security

Key Takeaways

Nozomi Networks joins Anthropic’s Project Glasswing to enhance AI-driven vulnerability detection in critical infrastructure.
The initiative aims to standardize vulnerability assessments across operational technology and connected systems.
Over 10,000 high-severity vulnerabilities have been identified in participants’ codebases using Claude Mythos Preview.

Collaboration for Cybersecurity

Nozomi Networks has partnered with Anthropic for Project Glasswing, leveraging AI to detect software vulnerabilities in critical infrastructure and connected systems. This collaboration focuses on enhancing security measures for operational technology (OT), the Internet of Things (IoT), and cyber-physical systems.

The initiative, which unites software developers, infrastructure providers, and cybersecurity experts, allows organizations to utilize Anthropic’s AI model, Claude Mythos Preview, to scan their codebases for vulnerabilities. So far, this model has identified over 10,000 potential flaws, categorized as high or critical severity, within the software of participating organizations.

Nozomi’s role extends the project’s outreach into software specifically designed for industrial applications, connecting enterprise services with physical systems. This integration highlights the need for security teams to evaluate vulnerabilities in the context of operational performance, safety, and reliability. Given that OT environments often include long-lifecycle equipment and limited maintenance windows, vulnerabilities must be prioritized based on their operational consequences.

NIST guidelines define OT as systems that interact with the physical environment and suggest that cybersecurity controls for these systems must address performance, reliability, and safety. Consequently, a technical severity rating is only part of the picture in prioritizing vulnerabilities; the affected software’s role in production processes and the potential impacts of failure must be assessed.

The Cybersecurity and Infrastructure Security Agency (CISA) also emphasizes the importance of visibility and categorization in vulnerability management. Organizations must identify critical functions affected by vulnerabilities before determining the urgency of a response. The high volume of findings generated by Project Glasswing necessitates careful human validation and prioritization based on operational significance.

Nozomi, already using AI to analyze industrial network communication and relationships, plans to integrate Project Glasswing’s models into its platform. However, the current information does not suggest that the project will involve live testing of customer equipment or production networks.

Dragos, another cybersecurity firm focused on industrial environments, joined Project Glasswing prior to Nozomi. The company is using Claude Mythos Preview to identify vulnerabilities within its own products, aiming to enhance both their software security and the overall understanding of AI performance in OT contexts. Findings from both companies will be shared with the broader cybersecurity community.

Testing without disrupting active operations is crucial. NIST recommends that organizations evaluate vulnerability scanning tools in controlled environments before applying them in real-time systems. Active testing can create traffic that affects OT components, potentially disrupting critical processes. Alternatives like simulated or virtualized systems can be employed for testing methods, ensuring real equipment remains unaffected.

Discovery and remediation of vulnerabilities are distinct phases. Identifying a flaw does not guarantee immediate patching, especially when maintaining system availability is a priority. The patch management process involves several steps, such as identifying vulnerabilities, prioritizing them based on operational needs, and ensuring that updates do not negatively impact functionality or safety.

Project Glasswing represents a significant endeavor in exploring how AI can help software developers, infrastructure operators, and security specialists collaboratively advance defensive vulnerability research. Nozomi is committed to contributing its insights and expertise to Anthropic’s broader efforts to pinpoint and mitigate weaknesses in critical software systems.

The content above is a summary. For more details, see the source article.