英国宣布 AI 风险包及扩展网络防御飞行员

条例

英国宣布 AI 风险包及扩展网络防御飞行员

在 Google 上的首选来源中添加 Unite.AI

英国政府宣布于2026年10月12日采取一揽子措施，保护英国免受日益强大的AI所带来的风险，包括扩展前沿-AI网络防御计划

内阁办公室、AI安全研究所、国家网络安全中心以及数字、文化、媒体和体育部发布了公告。 AI部长Kanishka Narayan阐述了伦敦帝国学院科技孵化器Base One的措施，他警告说，随着边境AI系统变得更加强大和能够自主行动，失败或恶意使用的潜在后果可能变得更加严重。

“我希望英国成为世界上最难使用人工智能攻击的国家之一，” 他承诺在必要时制定立法，并依靠英国的科学专长、技术能力和机构来了解日益强大的系统，应对新出现的威胁，并在出现错误时作出反应。 他的称公司建立世界上最强大的AI系统，应该注意科学家对严重危险的警告，调查出乎意料的行为，并扣留任何无法充分控制风险的系统。

根据公告，这项行动以两个飞行员为基础，这些飞行员帮助发现和解决了薄弱环节，并发现了潜在的网络威胁。 政府称将扩大该方案，帮助保护每天依赖的公共服务人员，在短期内减少风险，支持公共部门的长期收养。

网络法案变更和AI事件准备

政府称将探索加强《网络安全和复原力法案》的部分内容和更广泛的制度，以便更好地保护医院和公用事业等基本服务免受新出现的AI相关风险，包括AI失控风险，因为自主系统逃避人类监督。 它还将审查应如何查明和报告AI事件和接近失职的严重情况，以及如何迅速吸取经验教训以减少未来损害的风险，详细介绍任何措施，与议员、技术专家和需要提供这些措施的操作人员合作。

目前正与国家安全委员会和AISI合作，探讨如何应对涉及AI agents的事件，而AI agents是能够在很少人监督的情况下执行复杂任务的系统。 这种事件会比政府旨在处理的其他重大事件(称公告)更快地展开，并蔓延得更广，而迎接这一挑战将需要新的专门知识和能力。

该套件还延伸至AI验证，该公告称其是一个新兴领域，用于检查声称有关AI系统是如何开发，测试和评估的承受力检查。 联合王国政府称完全有能力在这一领域发挥主导作用，并将在今后几个月中详细说明支持联合王国AI核查工作的情况。

网络部长Liz Lloyd 称网络事件继续打乱公私部门大小组织，造成她所描述的在网络安全方面保持领先的商业迫切需要。 政府通过扩大AI保护公共服务的试点项目和推进网络法案，将网络安全和AI准备置于英国数字未来的核心。

脆弱性数字背后的试点

该公告引用了科学、创新和技术部(DSIT)和国家科学技术委员会(NCC)于2026年6月12日公布的一项案例研究，以支持其数字。 它记录了一个试点，其中政府网络协调中心(GC3)是国家通信服务委员会与DSIT之间的伙伴关系，每周进行面对面的黑客交易，利用边境AI扫描政府各部门的公共密码储存库。 各小组与AISI和NCCSC专家合作，设法找到和减轻以前未查明的弱点，然后才能加以利用。 GC3没有规定一种单一办法；各小组获得示范准入并建立了自己的工具。

案例研究描述了几个管道。 一个团队通过一个6级的AI agent链(triage，校验者，审计师，追踪器，法官及摘要)运行每个公共存储器，其中每个阶段读取并挑战最后一个，手动升级：一名团队成员检查每行，重新核实曝光，处理假阳性。 另一个团队首先运行了常规扫描工具，包括Gitleaks,Trivy,Semgrep和Hadolint，然后在顶部分层了三个模型阶段，覆盖发现，攻击链调查和分解。 第三个部门对数百项服务进行了全组织范围审计，将审计分为五个特定领域的Claude技能。

与会者总共确定了407项调查结果，包括使服务暴露于认证绕行、数据曝光和远程代码执行的关键弱点，案例研究报告。 所有重大弱点都得到了补救，没有发现任何发现的证据，该月9个政府组织花费13 000英镑。 作者称这些模型追踪了传统扫描仪无法做的跨服务界限的弱点，并将业务逻辑与技术细节联系起来。

一个关键的调查结果影响了遗留下来的GitHub行动，在一个支持关键政府数字服务的存储处。 对开放拉动请求进行专门设计的评论，可以让外部用户启动一系列工作流程，在GitHub Actions transfer上产生任意的远程代码执行，以及提取该工作流程可用的秘密和令牌的潜在路径。 GC3将启动第二阶段试点，更多部门，新增模式，并从公共代码延伸至封闭源码产业。

示范评价和更广泛的政府方案

Narayan证实，AISI科学家在与AI主要开发商和美国政府合作后，正在对新一代的高级AI模型进行严格的评价。

该公告还指向2026年4月15日公布的一封面向商界领袖的公开信，DSIT国务卿利兹·肯德尔(Liz Kendall)和保安部长丹·贾维斯(Dan Jarvis)在信中敦促企业认真对待其组织高层的网络安全，用政府的网络基本计划来掌握基本情况，并遵循NCSC的建议，包括签署其免费的预警服务，该服务提醒各组织注意潜在的网络攻击，以便在事件升级前采取行动。

根据公告，这些措施是漫长旅程中的第一步，总理的AI工作组正在率先开展更广泛的工作。 该公告指出首相9月在联合国大会上承诺尽一切可能保护英国人民免遭AI风险，称AI将成为英国20国集团主席的核心。 在今后几个月中，工作队将侧重于受大赦国际影响最大的领域，涉及国家安全、经济增长以及公共服务和劳动力的未来。

本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

Regulation

UK Announces AI Risk Package and Extends Cyber Defense Pilots

Add Unite.AI to your preferred sources on Google

The UK government announced a package of measures on October 12, 2026, to protect Britain from the risks posed by increasingly powerful AI, including extending a frontier-AI cyber defense program the government said has already helped find and fix more than 1,000 potential vulnerabilities across government systems.

The announcement was issued by the Cabinet Office, the AI Security Institute (AISI), the National Cyber Security Centre (NCSC) and the Department for Digital, Culture, Media and Sport. AI Minister Kanishka Narayan set out the measures at Base One, Imperial College London’s science and technology incubator, arguing for a whole-society approach to resilience. He warned that as frontier AI systems become more powerful and capable of acting autonomously, the potential consequences of failures or malicious use could become far more serious.

“I want Britain to become one of the hardest countries in the world to attack using artificial intelligence,” Narayan said. He pledged to legislate wherever necessary and to rely on British scientific expertise, technological capability and institutions to understand increasingly powerful systems, counter emerging threats and respond when things go wrong. He said companies building the world’s most powerful AI systems should heed their scientists’ warnings of serious dangers, investigate unexpected behavior and withhold any system whose risks cannot be adequately controlled.

According to the announcement, the move builds on two pilots that have helped find and fix the vulnerabilities and detect potential cyber threats. Extending the program, the government said, will help protect the public services people rely on every day, reduce risk in the short term and support longer-term adoption across the public sector.

Cyber Bill Changes and AI Incident Preparedness

The government said it will explore strengthening parts of the Cyber Security and Resilience Bill and the wider regime so that essential services such as hospitals and utilities are better protected from emerging AI-related risk, including AI loss-of-control risk, where autonomous systems evade human oversight. It will also examine how serious AI incidents and near misses should be identified and reported, and how lessons can be learned quickly to reduce the risk of future harm, with the detail of any measures worked through with parliamentarians, technical experts and the operators who would need to deliver them.

Work is also underway with the NCSC and AISI on how to respond to incidents involving AI agents, systems capable of carrying out complex tasks with little human supervision. Such an incident could unfold faster and spread wider than other major incidents governments are designed to handle, the announcement said, and meeting that challenge will require new expertise and capabilities.

The package also extends to AI verification, which the announcement describes as an emerging field that checks whether claims about how AI systems have been developed, tested and assessed withstand scrutiny. The government said the UK is well placed to take a leading role in the field and that more details on support for UK AI verification efforts will be set out in the coming months.

Cyber Minister Liz Lloyd said cyber incidents continue to disrupt organizations large and small across the private and public sectors, creating what she described as a business-critical need to stay on the front foot on cyber security. “Through scaling up projects piloting AI to protect public services and progressing a Cyber Bill to boost our defences, government is putting cyber security and AI preparedness at the heart of the UK’s digital future,” Lloyd said.

The Pilot Behind the Vulnerability Figures

The announcement cites a case study published on June 12, 2026, by the Department for Science, Innovation and Technology (DSIT) and the NCSC, in support of its figures. It documents a pilot in which the Government Cyber Coordination Centre (GC3), a partnership between the NCSC and DSIT, ran weekly in-person hackathons that used frontier AI to scan public code repositories across government. Working with AISI and NCSC specialists, teams sought to find and mitigate previously unidentified vulnerabilities before they could be exploited. GC3 did not mandate a single approach; teams received model access and built their own tooling.

The case study describes several pipelines. One team ran each public repository through a six-stage AI agent chain (triage, validator, auditor, tracer, judge and summary) in which each stage reads and challenges the last, with manual escalation: a team member checked every line, re-verified exposure and handled false positives. Another team ran conventional scanning tools first, including Gitleaks, Trivy, Semgrep and Hadolint, then layered three model stages on top covering discovery, attack-chain investigation and triage. A third department distilled an organization-wide audit across hundreds of services into five domain-specific Claude Skills.

Participants identified 407 findings in total, including critical weaknesses exposing services to authentication bypass, data exposure and remote code execution, the case study reports. All critical weaknesses were remediated and no evidence of exploitation was identified for any finding, at a cost of £13,000 in tokens across nine government organizations over the month. The authors said the models traced vulnerabilities across service boundaries, which traditional scanners cannot do, and linked business logic with technical detail.

One critical finding affected legacy GitHub Actions in a repository supporting a key government digital service. A specially crafted comment on an open pull request could let an external user set off a chain of workflows, producing arbitrary remote code execution on the GitHub Actions runner and a potential path to extract the secrets and tokens available to that workflow. GC3 will start a second phase of the pilot with more departments, additional models and an extension from public code to closed-source estates.

Model Evaluations and Wider Government Program

Narayan confirmed that AISI scientists are conducting rigorous evaluations of the latest generation of advanced AI models, following work with leading AI developers and the US government.

The announcement also points to an open letter to business leaders published on April 15, 2026, in which DSIT Secretary of State Liz Kendall and Security Minister Dan Jarvis urged businesses to take cyber security seriously at the top of their organizations, get the basics right with the government’s Cyber Essentials scheme, and follow NCSC advice, including signing up to its free Early Warning service, which alerts organizations to potential cyber attacks so they can act before an incident escalates.

According to the announcement, the measures are the first steps on a longer journey, and wider work is being spearheaded by the Prime Minister’s AI Taskforce. The announcement noted the Prime Minister’s commitment at the UN General Assembly in September to do whatever is necessary to protect British people from AI risks as they arise, and said AI will be at the heart of the UK’s G20 Presidency. In the coming months, the taskforce will focus on the areas most impacted by AI, spanning national security, economic growth and the future of public services and the workforce.

Auto-fetched and translated by AI Atlas. All rights belong to the original author.