Exclusive: Anthropic's plan to protect critical infrastructure

Anthropic has unveiled a new cybersecurity program aimed at safeguarding critical infrastructure sectors like power grids and water utilities. Recognizing the accelerating threat AI poses to these systems, Anthropic will equip participating organizations with its top-tier AI models, on-site engineers, and threat intelligence. The Critical Infrastructure Defense Program partners include major firms such as Accenture, Booz Allen Hamilton, CrowdStrike, and Nozomi Networks.

Anthropic's flagship offering, Claude, is being deployed to detect and patch vulnerabilities in third-party systems. Additionally, Anthropic is rolling out a free AI-powered vulnerability scanning service called OSS Scanner, which will periodically assess open-source software projects and generate automated reports containing potential weaknesses, their exploitation risks, and recommended fixes.

Notably, these reports will not be reviewed by human experts, potentially leading to inaccurate findings. Anthropic's strategy follows a pilot project called Project Glasswing, which granted select organizations access to its most capable AI to find security flaws. While the initiative demonstrates AI's potential to rapidly identify vulnerabilities, concerns persist about verifying and implementing fixes without disrupting critical infrastructure operations. The program's cost structure and safety protocols for deploying fixes remain unclear.