根据最新研究，世界上最先进的 AI 模型突然具备了为电网、水务系统等关键服务绘制网络攻击蓝图的能力，且成功率令人担忧。
The world's most advanced AI models are suddenly capable of blueprinting cyberttacks on critical services like power grids and water systems with alarming success rates, according to new research
为何重要：关键基础设施运营商所剩时间无几。网络安全公司 Frenos 的新研究显示，顶尖前沿模型在规划攻击方面表现出危险的熟练程度，而自由获取的 AI 模型让此类黑客攻击更易实施，只是时间问题。
Why it matters: Critical infrastructure operators are running out of time. Top-tier frontier models are dangerously adept at mapping out attacks, new research from cybersecurity company Frenos shows, and it's only a matter of time before freely available AI models make such hacks easier to execute.
“这是一个比我们以往见过的任何时钟都走得快的时钟，”美国电力公司前高级安全主管 Steve Swick 在简报会上对记者表示。
• "It's a clock that is moving faster than we've ever seen," Steve Swick, former top security executive at American Electric Power, told reporters during a briefing. Driving the news:
新闻驱动因素： Frenos 的研究人员测试了 10 个 AI 模型规划针对工业系统（如监测水处理过程的软件）攻击的能力。
Researchers at Frenos tested 10 AI models on their ability to plan attacks against industrial systems, such as the software that monitors water treatment processes.
- 10 个被测模型中有 3 个在大约 90% 的尝试中成功构建了攻击计划。
• Three out of 10 models tested successfully constructed attack plans in roughly 90% of attempts.
- 表现差异巨大：研究人员认定处于能力前沿的 3 个模型，成功率仅为 17.9% 至 31.9%。
• Performance varied widely: Three models the researchers considered to be at the frontier of capabilities succeeded in just 17.9% to 31.9% of attempts.
- 表现最强的外国开放权重模型成功率为 36.2%，而表现最强的美国开放权重模型成功率仅为 5.2%。
• The strongest foreign open-weight model succeeded 36.2% of the time, while the strongest U.S. open-weight model succeeded just 5.2%.
- Frenos 未披露被测模型的名称。
• Frenos did not disclose the names of the models tested.
深度观察此项研究包含近 1.7 万次测试，其中包括 2520 次构建完整攻击计划的尝试。
Zoom in The research included nearly 17,000 tests, including 2,520 attempts to construct complete attack plans.
- 模型获得潜在目标的信息，并被要求拼凑出实现特定目标所需的步骤。
• The models received information about potential targets and tasked with piecing together the steps needed to achieve specific objectives.
大局观：水务系统、制造工厂等关键服务长期以来一直是民族国家对手和网络犯罪分子的攻击目标。
The big picture: Critical services like water systems and manufacturing plants have long been targets for nation-state adversaries and cybercriminals.
- 许多运营商面临人员编制和预算限制。测试和升级其系统也可能引发其无力承担的基本服务中断风险。
• Many operators face staffing and budget constraints. Testing and upgrading their systems can also risk disruptions to essential services that they can't afford.
- 人名和地名一律使用中文字符进行转写；品牌名称和产品名称可保持原样（即使用拉丁字母）。
正文内容：
• Unlike traditional corporate networks, industrial systems can be sensitive to unexpected traffic and commands.
- 与传统的企业网络不同，工业系统对意外的网络流量和指令非常敏感。即使攻击尝试失败，也可能扰乱关键运营流程。
威胁等级： Frenos的联合创始人兼首席技术官Harry Thomas表示：“一旦攻击者入侵了工业网络，人工智能（AI）就能帮助他们悄无声息地分析系统结构并识别其中的薄弱环节，而人类黑客则受到时间限制。”
Even an unsuccessful attack attempt could disrupt essential operations. Threat level: Once attackers gain access to an industrial network, AI could help them quietly map out systems and identify weaknesses without the same time constraints facing human hackers, Frenos co-founder and CTO Harry Thomas said. • "AI doesn't have the problem with patience," Thomas said.
“AI没有耐心方面的问题，”Thomas说，“它不需要吃饭、睡觉，也不需要上厕所。”网络安全公司Armadin的运营技术主管Paul Shaver补充道：“AI代理已经能够执行侦察任务并分析大量数据，而且不会触发安全警报。”
"It doesn't need to eat. It doesn't need to sleep, and it doesn't need to use the restroom." • Paul Shaver, head of operational technology at cybersecurity firm Armadin, added that AI agents can already conduct reconnaissance and analyze large volumes of data without necessarily triggering security alerts.
研究细节： Thomas指出，研究人员使用了一套相对简单的工具来指导AI模型完成各项任务；如果使用更复杂的工具（这些工具能够提供更多背景信息和反馈），AI模型的性能可能会进一步提升。“更强大的工具本可以为这些AI模型带来更大的优势，”他说。
The intrigue Thomas said the researchers used a relatively simple setup to guide the models through their tasks. More sophisticated tools that provide additional context and feedback could improve their performance. • "A more capable harness could have given these AI models more of an edge," he said.
值得关注的趋势：最近几周，包括白宫、前沿AI企业以及网络安全组织在内的多个机构都推出了相关举措，旨在帮助降低关键基础设施运营的成本并提供所需资源。
What to watch Several initiatives — including from the White House frontier AI companies cybersecurity organizations— have cropped up in recent weeks to help lower the costs and provide resources for critical infrastructure organizations.