Anthropic Unveils AI Tools to Secure Critical Infrastructure and Open Source Projects

Anthropic announced a new security effort on October 8, 2026. It combines their artificial intelligence models with human experts to protect key utilities.

The Critical Infrastructure Defense Program starts with eleven outside companies already trusted by operators. These partners include consultants and vendors who manage safety for power grids and water systems.

Many industrial machines run for decades without stopping, so patches can be hard to apply immediately. Flaws might sit in the software for years before a chance to fix them arrives.

The new plan uses Claude models alongside onsite engineers to find risks first. They aim to help operators decide which updates are safe for their running systems.

A second part of the launch is a free scanner for open source projects. It will run periodic checks on eligible software code at no charge.

Reports from this scanner include a proof of concept and suggested patches for issues found. However, the company notes that some results may contain mistakes or wrong severity ratings.

Early testing showed that most findings were valid after expert review by penetration testers. But maintainers can still choose to get human verified reports instead of automated ones.

This work builds on a previous program called Project Glasswing which focused on verifying defenses. The costs for this scanner are covered by the Defender Advantage Fund established in August.

Some observers note that the commercial terms for partner access remain unclear at this time. It is not yet known if free model access applies to all groups involved.

Anthropic expects artificial intelligence will support defenders more strongly within two years. For now, finding flaws has become easier while repairing them still relies on human effort.