AI company moves to defend critical infrastructure and open-source projects from AI
Anthropic has launched an initiative to address vulnerabilities exposed by its own AI models, focusing on critical infrastructure and open-source projects. The company aims to provide defenders with tools currently being used by attackers, while also acknowledging that attackers may hold the advanta
Anthropic's Initiative to Counter AI-Driven Threats
Anthropic has initiated a program to help organizations patch vulnerabilities exposed by its AI models, particularly those related to critical infrastructure and open-source projects. This effort, part of the "Anthropic Cyber Mission," is a response to the growing concerns around AI's potential to be weaponized against infrastructure and software systems.
The initiative includes a Critical Infrastructure Defense Program, which combines Anthropic's advanced AI models with on-site engineers. This approach is designed to address the gap in in-house expertise for many organizations, leveraging partnerships with firms like Accenture, Booz Allen, and CrowdStrike. These partners are already engaged in safeguarding infrastructure against AI-driven threats.
Another component of the mission is the OSS Scanner, a service that provides free security scans for open-source projects. However, the service is not universally available and is limited to projects with significant impact on infrastructure and user security. Eligibility criteria mirror those used by Google for its OS-FUZZ project, ensuring that only established and influential projects are prioritized.
Anthropic acknowledges that AI models can be used to identify vulnerabilities, but the company also notes that attackers may have the upper hand for the next two years. According to Anthropic, AI systems have improved in detecting vulnerabilities, with LLMs now identifying over 85% of bugs, up from 20% at the start of 2025. However, the company emphasizes that defenders still face resource shortages and challenges in keeping pace with AI's evolving capabilities.
The initiative comes in the wake of recent incidents where AI models were used to breach organizations, including HuggingFace. These events have prompted Anthropic and other AI companies to reconsider their approach to oversight and security, leading to a greater emphasis on responsible AI deployment.
Anthropic's efforts highlight the dual role of AI in both creating and mitigating security risks. While the company is taking steps to address these challenges, it also recognizes that the balance of power between attackers and defenders may shift in the coming years as AI continues to evolve.
Source: The Register. Analysis based on publicly available information; no hands-on testing is claimed.
