A Canadian government just used Claude to hack itself before anyone else could

While most companies are still trying to figure out what to actually do with AI beyond writing emails, the Government of Alberta used it to scan 466 million lines of government code in 20 hours.

The Ministry of Technology and Innovation ran Claude Code, Anthropic's coding focused AI tool, across systems spanning 27 provincial ministries. Around 50 AI agents worked in parallel, hunting for security vulnerabilities, infrastructure weaknesses and gaps in documentation.

Finding problems was only step one. Where Claude Code spotted a vulnerability, it could generate a fix, test it and build it. If a system did not have automated tests to confirm a patch was safe, Claude wrote the tests first. Where old code was too outdated or messy to patch efficiently, Claude rebuilt it entirely in a more modern language.

Alberta did not stop at a one off clean up. The team built specialised Claude based agents for ongoing security work, including red team agents that actively probe applications for weaknesses and blue team agents that check compliance against security standards, plus separate agents reviewing code quality and public facing content.

Every patch still went through human engineers before it shipped. Nothing here was AI running unsupervised through government infrastructure.

The context makes this land harder. Alberta has been ramping up software spending after a genuine surge in cybersecurity incidents, and this is what that investment looks like in practice, not a chatbot demo, an actual government using AI to defend actual systems used by actual citizens.

This is worth paying attention to precisely because it is boring in the best way. No flashy launch video, no AI agent replaces your job headline, just a government quietly doing unglamorous infrastructure work faster than a human team could manage alone.

For anyone tired of AI hype pieces that never leave the demo stage, this is the opposite: a genuinely useful deployment, at real scale, with real human oversight built in. More of this, please, and fewer chatbots pretending to be strategists.