Anthropic Launches Cyber Mission to Aid Defenders with Tools and Research

Key Takeaways

• Anthropic has launched its “Cyber Mission” to enhance cybersecurity for critical infrastructure and open-source software.
• The initiative leverages advanced Claude AI models, threat research, and engineering expertise to identify and remediate vulnerabilities.
• Key components include the Critical Infrastructure Defense Program, partnering with 11 major security and industrial firms, and the free OSS Scanner for open-source projects.
• The mission prioritizes practical vulnerability remediation and aims to reduce exploitable weaknesses across vital sectors.

Anthropic officially initiated its Anthropic Cyber Mission on October 8, 2026, a strategic endeavor designed to bolster the defenses of critical infrastructure and open-source software. This comprehensive initiative integrates Anthropic’s sophisticated Claude AI models with dedicated engineering talent, cutting-edge threat research, and substantial funding. A core tenet of the mission is to actively facilitate the remediation of security flaws, moving beyond merely identifying vulnerabilities to implementing tangible fixes.

The launch introduces two primary components: the Critical Infrastructure Defense Program and the OSS Scanner, a complimentary service available to participating open-source projects. Anthropic states that this mission builds upon insights gained from its earlier Project Glasswing. While Project Glasswing accelerated bug discovery, it highlighted a significant challenge: the subsequent processes of validating findings, prioritizing risks, and deploying patches remain resource-intensive and demand skilled human intervention.

Protecting Critical Infrastructure

The Critical Infrastructure Defense Program deploys Claude models, on-site engineers, and specialized threat research to trusted providers managing essential services such as power grids, water utilities, manufacturing facilities, transportation networks, and government systems. This program counts 11 foundational partners, including industry leaders like Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation.

These organizations are critical to supporting operational technology (OT), which encompasses the industrial controllers, control software, and networks that govern physical equipment. Many OT systems are designed for decades-long operation and cannot be easily taken offline for routine updates. The potential for a poorly tested change to disrupt production or vital services makes patch deployment in these environments far more complex than in typical enterprise IT networks.

Anthropic reports that several partners are already leveraging Claude models to identify and help customers address vulnerabilities. This initial consortium will evaluate which AI-assisted approaches prove effective and safe within live operational environments. The company acknowledges that AI alone cannot resolve every security challenge inherent in infrastructure, particularly given the constraints imposed by legacy equipment, stringent maintenance schedules, and paramount operational safety requirements.

Furthermore, Anthropic’s government defense program has extended Claude models and technical assistance to over half of U.S. states since June. This support covers various public sector security tasks, including code scanning, patching efforts, and incident response.

Free Open-Source Vulnerability Scanning

The OSS Scanner provides regular security scans utilizing Anthropic’s most advanced models. Open-source project maintainers must opt in to receive these reports, which include a detailed explanation of the vulnerability, a proof of concept demonstrating its exploitability, and a proposed patch when available. Unlike Anthropic’s established coordinated vulnerability disclosure process, these reports are delivered without human review.

This streamlined delivery method accelerates the process but places the responsibility on maintainers to verify the accuracy and impact of the findings. Anthropic anticipates a true-positive rate exceeding 90%, although it notes that severity ratings may occasionally be incorrect. Projects lacking the internal capacity to review raw findings will continue to receive human-verified reports through Anthropic’s traditional coordinated vulnerability disclosure process.

The underlying research for the OSS Scanner underscores the significance of this change. Anthropic’s data indicates over 29,000 candidate vulnerabilities identified during six months of scanning, yet only approximately 6,000 underwent manual review and triage. It is crucial to understand that these candidate findings should not be considered confirmed flaws or completed fixes without further validation.

Earlier reporting on Anthropic’s expanded Project Glasswing highlighted a shift toward moving beyond mere detection into active patching and other defensive actions. This new mission deepens that commitment by combining access to AI models with dedicated engineering support and resources for maintainers.

Project Glasswing has also been integrated into the broader Cyber Verification Program. This program features three access tiers, differentiating between defensive work, authorized penetration testing, and restricted testing for safety-critical systems. Applicants are subject to specific verification requirements and security controls tailored to their scope of work.

The Defender Advantage Fund provides financial backing for pilot projects and ensures the OSS Scanner remains a free service. Anthropic’s future plans include expanding its infrastructure partnerships, refining automated triage and patching capabilities, and conducting research into inherently safer software designs. The overarching goal is to deliver practical protection, quantifiable by a reduction in exploitable weaknesses, the reliability of essential services, and quicker recovery times following successful attacks.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.