美国海军部发布"AI优先"舰队战略，将慢速采用视为比"不完美对齐"更大的风险

海军部希望打造一支“AI优先”的舰队。一项新战略阐述了如何更快地将数据和人工智能转化为战场优势。

负责监督海军和海军陆战队的海军部，已正式批准了新的“数据与人工智能武器化战略”。代理海军部长洪·曹（Hung Cao）签署了该文件，使其立即生效。该战略在海军部首席数据与人工智能官的领导下，经过一年多的开发，并与海军和海军陆战队的AI专家合作制定完成。

曹表示，该战略将使海军部通过快速部署数据和AI，“在学习和作战方面领先任何对手”。他将其描述为构建“AI优先”舰队的路线图，将信息转化为军事优势，并实现更快、更好的决策。

战略的核心是“Bits2Effects循环”，这是一个五阶段的数字化适应框架。它追踪从自动收集军事数据，到传输、分类和分析，再到将其用于实际军事决策和行动的路径。所获得的经验教训会反馈到循环中，使系统、战术和训练持续更新。

关键指标是“平均效果时间”（Mean Time to Effect, MTTE），用于衡量从捕获新数据到产生具体军事响应或适应所需的时间。窗口越短，部队反应和调整的速度就越快。在经历多轮学习周期的持久冲突中，根据该战略文件，学习和适应最快的部队将占据主导地位：https://www.doncio.navy.mil/ContentView.aspx?id=20519。

公告：https://www.navy.mil/Press-Office/Press-Releases/display-pressreleases/Article/4545237/acting-secretary-of-the-navy-signs-strategy-to-weaponize-data-and-artificial-in/ 列出了六个目标：加快作战AI部署，提高数据可用性和可用程度，扩展技术基础设施，简化审批流程，加强人员的数据和AI素养，并加深与产业界、学术界、政府机构及盟友的合作。

这些措施中许多计划在2027财年的第一季度之前落实，该财年于2026年12月结束。到2029财年末，合格的数据工程师、数据科学家以及人工智能和机器学习工程师的人数预计将翻一番。

该战略计划直接在军舰和海军陆战队远征单位上运行大型语言模型和具代理性的人工智能。这些系统即使在通信被干扰或切断的情况下也需要正常工作。军人将会在这些系统基础上开发自己的应用程序。一个“人工智能战争委员会”将优先确定使用案例、协调资源，并预先批准战争期间的数据共享、分类和部署规则的变更。

该战略文件采用了国防部更广泛人工智能战略中一个特别深远的权衡：行动过慢的风险比这些系统不完美对齐的风险更大。这一表述出现在“战争时期方法”的背景下。国防部希望像国家已处于战时一样处理风险评估和组织障碍，做出有利于速度的决策。

海军的这一战略是美国武装部队更广泛人工智能转型的一部分，Business Insider：https://www.businessinsider.com/us-navy-data-ai-for-ai-first-fleet-2026-7 报道称。GenAI.mil 是一个中央平台，供国防部人员和雇员使用生成式人工智能，其日活跃用户在2026年6月达到150万，而2025年12月上线时仅为8万。其用途覆盖从日常办公任务到军事计划和作战行动。

陆军正在测试一种“下一代指挥与控制”系统中的人工智能，以更快处理大量数据，并帮助士兵建立态势感知和做出决策。据报道，海军的一个人工智能项目将潜艇计划任务从160小时缩短至十分钟。

这些应用的真实性在对伊朗的战争中变得清晰。据报道，美国军方使用：https://the-decoder.com/us-military-uses-anthropics-claude-for-ai-driven-strike-planning-in-iran-war/ Anthropic的语言模型Claude进行目标分析和打击计划。该部署具有高度政治性。

特朗普政府在安thropic要求对完全自主武器和大规模国内监控实施限制后，将其锁定在政府系统之外：https://the-decoder.com/the-pentagon-openai-anthropic-fallout-comes-down-to-three-words-any-lawful-use/。不久之后，OpenAI 与五角大楼达成协议：https://the-decoder.com/anthropic-ceo-attacks-openais-pentagon-deal-as-safety-theater-while-investors-scramble-for-de-escalation/，在机密网络上运行其模型。OpenAI 也提出了类似的红线，但依赖合同和技术保障，而不是强制政策要求。海军的新战略可能会进一步推高军方对强大语言模型和 AI 代理的需求。

AI 军备竞赛正在全球范围内展开。中国正在快速推动军事 AI 的采用。乔治城大学的研究人员分析了数千份公开可获得的中国人民解放军采购请求：https://the-decoder.com/thousands-of-procurement-documents-show-how-chinas-army-wants-to-weaponize-ai/。这些文件显示，北京正在测试用于无人作战车辆、网络防御、舰船追踪、陆地、海上及太空目标获取以及深度伪造驱动的虚假信息的 AI 系统。

北约也已经在实际操作中使用 AI：https://the-decoder.com/u-s-military-strikes-3000-targets-in-iran-with-ai-support-but-oversight-remains-underinvested/。法国海军上将皮埃尔·万迪埃（Pierre Vandier），北约数字化转型的最高官员，表示盟国正在用 AI 追踪俄罗斯的影子油轮舰队。以色列在对抗伊朗战争前花了数年时间部署 AI，从海量拦截情报数据中筛选信息：https://the-decoder.de/wsj-bericht-liefert-neue-details-zum-ki-einsatz-im-krieg-gegen-den-iran/。

在美国方面，五角大楼正大力投资于整合商业 AI：https://the-decoder.com/eight-tech-giants-sign-pentagon-deals-to-build-an-ai-first-fighting-force-across-classified-networks/，并计划进一步让 AI 公司在机密数据上训练军用专用模型：https://the-decoder.com/pentagon-plans-to-let-ai-companies-train-models-on-classified-data/。这将是一次质的飞跃。敏感情报将直接嵌入模型中。

网络安全的节奏正在迅速加快。与核升级的类比不再只是比喻。中国网络安全公司奇虎360的创始人周鸿祎明确提出了这一比较：https://the-decoder.com/chinese-cybersecurity-firm-builds-ai-tools-to-rival-mythos-and-frames-the-race-as-cyber-nuclear-deterrence/。他认为，像Anthropic的Claude Mythos这样的AI模型能够自主发现漏洞并构建攻击链，相当于“AI时代的网络核武器”。

这种言辞背后的紧迫性基于可衡量的技术进展。英国人工智能安全研究所：https://the-decoder.com/new-claude-mythos-becomes-the-first-ai-model-to-clear-all-cyberattack-simulations-from-britains-ai-safety-agency/ 修正了对AI网络能力翻倍速度的估计，仅几个月内就上调了两次。美国政府现在将这些模型视为战略资产，并最初阻止Anthropic公开推出其Fable 5 AI模型：https://the-decoder.com/us-government-forces-anthropic-to-disable-claude-fable-5-and-mythos-5-for-all-customers-worldwide/。

周称Fable 5是“神话系统的民用、阉割版”，即Anthropic最强的网络安全模型：https://the-decoder.com/anthropic-gets-us-approval-to-bring-back-claude-mythos-5/，并指出美国担心外国行为者会越狱该系统以达到Mythos水平的能力。周说：“这是美国政府最不能容忍的事情。它必须确保只有自己拥有这种能力，从而对这一战略资产形成绝对垄断。”

欧盟处于边缘状态：https://the-decoder.com/anthropic-shutdown-sparks-sovereignty-debate-across-europe/，依赖大型美国科技公司的善意，因为并不存在可比的欧洲产品：https://the-decoder.com/claude-mythos-is-a-wake-up-call-for-europes-ai-safety-apparatus/。

保持对AI的了解。清晰、有用，无废话。

关注The Decoder，获取AI新闻、背景故事和专家分析。

The Decoder：https://the-decoder.com/

The Department of the Navy wants to build an "AI-first" fleet. A new strategy lays out how to turn data and AI into battlefield advantages faster.

The Department of the Navy, which oversees both the Navy and the Marine Corps, has formally approved a new "Strategy to Weaponize Data and Artificial Intelligence." Acting Secretary of the Navy Hung Cao signed the document, putting it into effect immediately. The strategy took more than a year to develop under the department's Chief Data and Artificial Intelligence Officer, working with AI experts across the Navy and Marine Corps.

Cao said the strategy would let the Department of the Navy "out-learn and out-fight any adversary" through rapid deployment of data and AI. He described it as a roadmap for building an "AI-first" fleet that turns information into military advantage and enables faster, better decision-making.

At the heart of the strategy is the "Bits2Effects Cycle," a five-stage framework for digital adaptation. It traces the path from automated collection of military data through transmission, classification, and analysis to its use in real military decisions and actions. Lessons learned feed back into the cycle, allowing continuous updates to systems, tactics, and training.

The key metric is "Mean Time to Effect," or MTTE. It measures how long it takes from the moment new data is captured until it produces a concrete military response or adaptation. The shorter that window, the faster a force can react and adjust. In a drawn-out conflict with multiple learning cycles, the force that learns and adapts fastest will dominate, according to the strategy paper：https://www.doncio.navy.mil/ContentView.aspx?id=20519.

The announcement：https://www.navy.mil/Press-Office/Press-Releases/display-pressreleases/Article/4545237/acting-secretary-of-the-navy-signs-strategy-to-weaponize-data-and-artificial-in/ lays out six goals: speed up operational AI deployment, improve data availability and usability, expand technical infrastructure, streamline approval processes, strengthen data and AI literacy among personnel, and deepen collaboration with industry, academia, government agencies, and allies.

Many of these measures are supposed to be in place by the first quarter of fiscal year 2027, which ends in December 2026. By the end of fiscal year 2029, the number of qualified data engineers, data scientists, and AI and machine learning engineers is supposed to double.

The strategy calls for running large language models and agentic AI directly on warships and with Marine Corps expeditionary units. These systems need to work even when comms are jammed or cut off. Service members would build their own apps on top of them. An "AI War Council" would prioritize use cases, coordinate resources, and pre-approve wartime changes to data sharing, classification, and deployment rules.

The strategy paper adopts a particularly far-reaching trade-off from the Department of Defense's broader AI strategy: the risks of moving too slowly outweigh the risks of imperfect alignment in these systems. That passage sits within the context of a "Wartime Approach." The department wants to handle risk assessments and organizational hurdles as if the country were already at war, making decisions that favor speed.

The Navy's strategy is part of a broader AI transformation across the US armed forces, Business Insider：https://www.businessinsider.com/us-navy-data-ai-for-ai-first-fleet-2026-7 reports. GenAI.mil, the central platform where Defense Department personnel and employees can use generative AI, hit 1.5 million daily users in June 2026. That's up from 80,000 when it launched in December 2025. Uses range from routine office tasks to military planning and combat operations.

The Army is testing AI in a "Next Generation Command and Control" system to process large volumes of data faster and help soldiers build situational awareness and make decisions. A Navy AI program reportedly cut a submarine planning task from 160 hours down to ten minutes.

How real these applications already are became clear during the war against Iran. The US military reportedly used：https://the-decoder.com/us-military-uses-anthropics-claude-for-ai-driven-strike-planning-in-iran-war/ Anthropic's language model Claude for target analysis and strike planning. The deployment is politically charged.

The Trump administration locked Anthropic out of government systems：https://the-decoder.com/the-pentagon-openai-anthropic-fallout-comes-down-to-three-words-any-lawful-use/ after the company insisted on restrictions for fully autonomous weapons and mass domestic surveillance. Shortly after, OpenAI struck a deal with the Pentagon：https://the-decoder.com/anthropic-ceo-attacks-openais-pentagon-deal-as-safety-theater-while-investors-scramble-for-de-escalation/ to run its models on classified networks. OpenAI cites similar red lines but relies on contractual and technical safeguards rather than hard policy demands. The Navy's new strategy is likely to push military demand for powerful language models and AI agents even higher.

The AI arms race is playing out on a broad scale worldwide. China is pushing military AI adoption at a rapid clip. Researchers at Georgetown University analyzed thousands of publicly available procurement requests：https://the-decoder.com/thousands-of-procurement-documents-show-how-chinas-army-wants-to-weaponize-ai/ from the Chinese People's Liberation Army. The documents show Beijing is testing AI systems for unmanned combat vehicles, cyber defense, ship tracking, target acquisition on land, at sea, and in space, and deepfake-powered disinformation.

NATO is already using AI operationally：https://the-decoder.com/u-s-military-strikes-3000-targets-in-iran-with-ai-support-but-oversight-remains-underinvested/ as well. French Admiral Pierre Vandier, NATO's top officer for digital transformation, said alliance members are using AI to track Russia's shadow tanker fleet. Israel spent years deploying AI to sift through the flood of intercepted intelligence data ahead of its war against Iran：https://the-decoder.de/wsj-bericht-liefert-neue-details-zum-ki-einsatz-im-krieg-gegen-den-iran/.

On the US side, the Pentagon is investing heavily in integrating commercial AI：https://the-decoder.com/eight-tech-giants-sign-pentagon-deals-to-build-an-ai-first-fighting-force-across-classified-networks/ and plans to go further by letting AI companies train military-specific model versions on classified data：https://the-decoder.com/pentagon-plans-to-let-ai-companies-train-models-on-classified-data/. That would be a qualitative leap. Sensitive intelligence would be baked directly into the models.

The pace is picking up fast in cybersecurity. The parallels to nuclear escalation aren't just a metaphor anymore. Zhou Hongyi, founder of Chinese cybersecurity firm Qihoo 360, drew the comparison explicitly：https://the-decoder.com/chinese-cybersecurity-firm-builds-ai-tools-to-rival-mythos-and-frames-the-race-as-cyber-nuclear-deterrence/. He argued that the ability of AI models like Anthropic's Claude Mythos to autonomously find vulnerabilities and build attack chains amounts to "cyber nuclear weapons of the AI age."

The urgency behind that rhetoric is grounded in measurable technical progress. The UK's AI Security Institute：https://the-decoder.com/new-claude-mythos-becomes-the-first-ai-model-to-clear-all-cyberattack-simulations-from-britains-ai-safety-agency/ revised its estimate for how fast AI cyber capabilities are doubling, adjusting it upward twice in just a few months. The US government now treats these models as strategic assets and initially blocked Anthropic from publicly launching its Fable 5 AI model：https://the-decoder.com/us-government-forces-anthropic-to-disable-claude-fable-5-and-mythos-5-for-all-customers-worldwide/.

Zhou called Fable 5 a "civilian, neutered version of Mythos," Anthropic's most capable cybersecurity model：https://the-decoder.com/anthropic-gets-us-approval-to-bring-back-claude-mythos-5/, and suggested the US feared foreign actors would jailbreak the system to reach Mythos-level capabilities. "This is what the US government finds most intolerable. It must ensure that it alone possesses this capability, forming an absolute monopoly over this strategic asset," Zhou said.

The European Union is stuck on the sidelines：https://the-decoder.com/anthropic-shutdown-sparks-sovereignty-debate-across-europe/, dependent on the goodwill of big US tech companies because comparable European products don't exist：https://the-decoder.com/claude-mythos-is-a-wake-up-call-for-europes-ai-safety-apparatus/.

Stay in the loop on AI. Clear, useful, no fluff.

Follow The Decoder for AI news, background stories and expert analyses.

The Decoder：https://the-decoder.com/