Signal Brief

Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities across government systems

阿尔伯塔省政府技术与创新部使用Claude Code(Opus和Sonnet模型)在20小时内扫描了4.66亿行代码,修复了安全漏洞,并构建了持续安全审查智能体。传统方法估计需要6.5年,而重建25年旧的Java门户仅用4-5天。

Anthropic-news 发布 2026-07-06 收录 2026-07-06 观察

一句话判断

展示了Claude在政府大规模代码审计中的实际效率数据(传统6.5年 vs 20小时)和重建老旧系统的具体案例(25年旧Java系统4-5天重建)。

核心信息

阿尔伯塔省政府技术与创新部使用Claude Code(Opus和Sonnet模型)在20小时内扫描了4.66亿行代码,修复了安全漏洞,并构建了持续安全审查智能体。传统方法估计需要6.5年,而重建25年旧的Java门户仅用4-5天。

原始内容

![Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities across government systems](https://www-cdn.anthropic.com/images/4zrzovbb/website/802260d34a0653f23fd4944fae43064df367aa44-1000x1000.svg) Since 2025, the Government of Alberta has been using Claude Code with both Opus and Sonnet models to review its systems, find vulnerabilities, and fix them. A team inside Alberta’s Ministry of Technology and Innovation scanned 466 million lines of code in 20 hours, remediated security gaps across its systems, and built new tools to make those systems safer. We’re sharing details of their experience as an example of how government agencies can use Claude and Claude Code to secure their systems at a large scale. This is a critical challenge, as governments rely on these systems to deliver benefits and keep services running—yet the code is often old, insecure, and incompletely documented. Alberta has also published a collection of technical white papers documenting its efforts for other governments to learn from; [you can read them here.](https://thevelocitywhitepapers.com/) “Albertans trust their government with some of the most sensitive information in their lives, and it is our responsibility to protect it,” said Nate Glubish, Alberta’s Minister of Technology and Innovation. “By using AI to find and fix vulnerabilities across our systems, we accomplished in hours what would have taken a traditional approach years to complete. This is what responsible government looks like in the AI era, and the best is still ahead of us.” Alberta’s approach ------------------ Alberta’s Ministry of Technology and Innovation maintains the systems of all 27 provincial ministries, from social services to public safety to wildfire response. That includes roughly 1,280 applications and 3,400 code repositories. Most of it has never undergone a systematic security review, and the accumulated technical debt—insecure code, unaddressed bugs, outdated software—runs into the billions of dollars. The Ministry’s systems hold highly sensitive information, including tax records, government procurement data, and social services case files. So in 2025, the Ministry set up an internal team with a mandate to make these systems more secure and easier to maintain over time, working with Claude to do so. Already, the Ministry has used Claude to: **Assess 466 million lines of government code in 20 hours.** The team put Claude to work on the codebases it maintains, using Claude Code with Claude Opus and Sonnet models. Around 50 agents worked autonomously and in parallel to scan the systems for security vulnerabilities, weaknesses in underlying infrastructure and deployment processes, and gaps in technical documentation. Claude Code ran a two-stage routine, first scanning each repository with a rules engine to flag known patterns, then reviewing those flags and citing the exact file and line for each finding so developers could verify them. The scan covered every repository Alberta owns and identified issues that traditional automated scanning tools had missed. It took about 20 hours for Alberta’s implementation; the team estimates that that kind of review could otherwise have taken around 6.5 years. **Fix the vulnerabilities the scan found.** Where the scan identified a vulnerability, Claude Code could often generate a fix, test it, and build it. In cases where a system lacked the automated tests needed to confirm that a patch was safe, Claude wrote the tests first. Where the code was too outdated or too complex to patch efficiently in its existing form, Claude rebuilt it in a more modern and maintainable language. In some scenarios, these systems could be rebuilt in as little as four to five days, including a subsidy program portal that was originally hand-coded in Java roughly 25 years ago and took five months to build the first time. All of this was done in partnership with the Ministry’s engineers: before any patch shipped, it was reviewed and approved by the team. **Run continuous security review.** Alberta’s cybersecurity team also built a set of specialized Claude review agents that run throughout the development process. A “red team” agent probes an application from the outside, the way an attacker might, and maps how a vulnerability might be exploited. A “blue team” agent then assesses the application’s defenses against an international security standard, and writes a remediation plan that points to the exact files to fix. Additional agents check code quality and the clarity of the writing the public sees. Every application is checked against roughly 95 security controls on each pass. These agents are built on top of the [Claude Agent SDK](https://code.claude.com/docs/en/agent-sdk/overview) and run a robust series of checks and analysis for every application. In addition to scanning, securing, and modernizing its own systems, Alberta is training both government workers and the public in the use of AI through the Alberta AI Academy. Thousands of government employees and more than 10,000 members of the public have used the platform to learn the essentials of effective AI use, from prompting through enterprise application delivery. Through the Academy, the Ministry of Technology and Innovation aims to extend its approach beyond a single team or project to every ministry that needs it. Looking ahead ------------- Today, Claude helps the Ministry write, review, and deploy code that aids in its modernization efforts. Next, it plans to expand that work with AI agents that can build entirely new software and tools alongside engineers. The Government of Alberta also plans to continue its modernization work. One ministry, for example, has 185 legacy applications running in production, which are expensive to maintain and difficult to update. The Government plans to use Claude Code to analyze these systems, understand what they do, and consolidate them into 16 reusable applications built on modern coding languages and conventions. The goal is to reduce complexity, lower maintenance costs, and speed up modernization work that would otherwise take years to complete. A case study for governments ---------------------------- The technical debt and security vulnerabilities the Government of Alberta is working to address are hardly unique. They exist in the systems of many provinces, states, and federal agencies across the world. The [technical white papers](https://thevelocitywhitepapers.com/) Alberta has released give other governments a blueprint for addressing these same issues. In addition to the white papers, Alberta is hosting [an industry day](https://luma.com/yzd00tir) in Edmonton in July to share what it has learned. And this fall, it will begin a program to scale its approach across the provincial government. We’ll keep working with Alberta as it expands these efforts, and we hope the approach it has documented can help other governments secure their own systems.

相关动态

02

Kimi K3 在 SpreadsheetBench 2 上排名第一

Kimi K3 在 SpreadsheetBench 2 基准测试中排名第一,超越 Claude Fable 5,完成 34.8% 的 workflow 任务。该基准测试涉及 321 个专家策划任务,平均每个任务包含 11.8 个工作表和 593.5 个单元格更改,聚焦于完整的电子表格工作簿执行。

twitter关注列表2026-07-18#模型#技术突破#评测
观察
03

Mind Lab 开源长文本强化学习项目

Mind Lab 开源了一个长文本强化学习(RL)项目,支持 2M tokens 的长上下文。相比以往需要数千个 GPU 的 1M 上下文研究,该项目仅需 8 个 GPU 即可完成,大幅降低了超长上下文研究的算力门槛。

twitter关注列表2026-07-17#开源#技术突破#模型
观察
04

Kimi K3非对称竞争分析

Kimi K3在Arena前端/WebDev人类盲测中以1679 Elo排名第一,领先Fable 5(1631)和GPT-5.6 Sol(1618),但在综合智能指数中仅排第四(57.1分),落后Fable 5(59.9)和GPT-5.6 Sol(58.9)。K3定价15美元/百万输出tokens,远低于Fable 5的50美元,并计划7...

twitter关注列表2026-07-18#AI#模型#技术突破
观察
05

二年级学生Jo Nagai发现蝴蝶可遗传记忆

日本东京都二年级学生Jo Nagai注意到养育的食蚜 caterpillars 在蝴蝶化后仍保持对薰衣草的回避行为,经Georgetown大学Entomologist Dr. Martha Weiss合作完成实验:70%训练过的蝴蝶及其后代均表现出对薰衣草的遗传性回避,记忆在全变态发育中保存并遗传。

twitter关注列表2026-07-18#研究#技术突破#信息
值得跟进