Artificial Intelligence

Anthropic Fast-Tracks AI Bug Reports and Taps OT Firms

Anthropic has rolled out two major cybersecurity initiatives aimed at accelerating open source bug reporting and bolstering operational technology defenses.

Radwan AbdulhadiEditor and technology writer

Published: October 9, 2026 · Updated: October 9, 2026 · 3 min read

Anthropic Fast-Tracks AI Bug Reports and Taps OT Firms — Image: SecurityWeek

Introduction to Anthropic's New Cybersecurity Initiatives

Anthropic announced a pair of new cybersecurity initiatives designed to address vulnerabilities in open source software and critical operational technology systems. According to details shared in reports from SecurityWeek, these programs build directly on lessons learned from Project Glasswing.

While finding software vulnerabilities has become significantly easier with modern AI capabilities, the company noted that verifying, prioritizing, and patching those flaws remains a difficult challenge. Flaws uncovered during Project Glasswing frequently required months to achieve proper remediation.

Launching the Unreviewed OSS Scanner

Inspired by Google’s OSS-Fuzz initiative, Anthropic has launched the OSS Scanner as a free service that utilizes its most capable models to periodically scan open source projects. Open source maintainers must explicitly opt in to have their codebases reviewed by the automated system.

Each generated report details the potential vulnerability, provides a proof-of-concept (PoC) demonstrating how it can be exploited, and suggests a fix when available. Responding to requests from maintainers who handle triage at scale, Anthropic elected to bypass human review entirely to speed up delivery.

As noted by the AI giant, the reports are model-generated and sent without human review. While this skips human bottlenecks, it may occasionally introduce inaccuracies such as incorrect severity ratings. Anthropic maintains a target true-positive rate above 90% and intends to refine the system continuously.

Projects that lack the capacity to process rapid model-generated alerts will continue to receive traditional, human-verified disclosures through Anthropic’s coordinated vulnerability disclosure channels. Further background on the company's broader safety and access frameworks can be found via Anthropic resources.

Targeting Operational Technology Security

In addition to open source software, Anthropic is turning its attention to industrial environments through the Critical Infrastructure Defense Program (CIDP). This effort integrates frontier Claude models, on-site engineers, and proprietary threat research to support operators in power, water, manufacturing, and transportation sectors.

The program brings together a select group of founding partners to help secure operational technology. Comprehensive details regarding the mission and strategic deployment of these tools were outlined when Anthropic said the initiative aims to protect essential services.

Founding Partners in the Critical Infrastructure Defense Program

The initial cohort for the critical infrastructure program consists of 11 prominent firms spanning consulting, technology, security, and industrial manufacturing. The founding partners include Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation.

These organizations help build, secure, and patch industrial equipment relied upon by vital utilities. Anthropic emphasized that operational technology systems present unique hurdles, often making it impossible to take systems offline for updates and leaving known vulnerabilities unresolved for extended periods.

Addressing Long-Term Patching Challenges

Operational technology environments frequently operate under strict uptime requirements. According to Anthropic, applying a security patch in rare cases can take decades to accomplish safely within complex industrial setups.

Several founding partners are already utilizing Claude models to identify and fix existing vulnerabilities while assisting their own customers with remediation efforts. Anthropic is intentionally launching the program with a small group of providers to measure practical effectiveness before expanding to additional sectors.