Government of Alberta Scans 466 Million Lines of Code With Claude in 20 Hours.

The Government of Alberta has completed a sweeping security audit of its critical infrastructure systems using Claude Code, scanning 466 million lines of code in approximately 20 hours to identify and remediate vulnerabilities.

The project, announced on LinkedIn by Anthropic, represents one of the first large-scale deployments of AI-assisted code scanning in a government context. Alberta's IT teams used both Opus and Sonnet models through Claude Code to analyze systems that support critical infrastructure and public services.

The scope of the audit underscores the challenge government agencies face in maintaining secure systems. Scanning nearly half a billion lines of code manually would require months of work from security teams. Claude Code compressed that timeline to under a day, identifying security gaps that could have left critical systems vulnerable to attack.

Beyond detection, the project included active remediation. Alberta's teams used Claude to fix identified vulnerabilities and build new internal security tools designed to prevent similar issues from emerging in the future. The official blog post provides additional technical details on the implementation.

This follows Anthropic's recent work with government agencies on security initiatives. The company recently restored Fable 5 access after a government cybersecurity review, and has been expanding its Project Glasswing program to help organizations implement AI safely.

The Alberta deployment demonstrates how AI code analysis tools can address resource constraints in government IT departments. Many agencies operate legacy systems with large codebases but limited security staff. Automated scanning tools like Claude Code can augment human reviewers, surfacing vulnerabilities that might otherwise go undetected for years.

However, the project also raises questions about model access and security review processes for government deployments. Anthropic has emphasized safety guardrails in recent releases, and government use cases typically require additional scrutiny around data handling and model behavior.

The collaboration marks a shift in how public sector organizations approach code security. Rather than periodic manual audits, AI-assisted scanning enables continuous monitoring of large codebases, potentially catching vulnerabilities before they can be exploited. For Alberta, the 20-hour audit cycle means security reviews that once took quarters can now happen in less than a day.