内容来源：https://aibusiness.com/generative-ai/the-more-operational-ai-becomes-bigger-security-challenge

内容总结：

AI安全进入新阶段：企业从“被动防御”转向“主动加固”

随着生成式AI加速落地，企业正面临前所未有的安全挑战。谷歌云、OpenAI等科技巨头本周密集发布新举措，标志着行业正从单纯保护模型与数据，转向构建可自主运行、跨系统交互的“操作型AI”安全体系。

OpenAI启动“黎明”计划，构建生态防线
OpenAI本周推出名为“Daybreak”的网络安全计划，旨在从设计阶段为AI系统注入抗风险能力，而非仅依赖事后补救。该计划已获思科、CrowdStrike、Cloudflare等多家安全厂商参与，反映出AI安全正从单点防御走向生态协同。

安全困境：AI既是武器也是目标
报道指出，旨在提升效率的AI系统正创造出新的攻击面、治理盲区与运营风险。近期教育平台Canvas遭攻击事件警示：许多企业尚未管理好现有互联系统，却又仓促接入更多AI，使安全漏洞雪上加霜。

部署瓶颈：人才与基础设施跟不上AI速度
谷歌云正大力招募AI部署工程师，OpenAI也独立成立咨询业务，凸显企业级AI落地的现实困难——组织采用AI的速度远超员工培训进度，而监控、治理与基础设施能力在愈发复杂的智能体环境中捉襟见肘。业界焦点已从“如何构建模型”转向“如何安全运营大规模自主系统”。

【行业动态速览】

• AI重塑企业运营：传统监控工具被迫升级，适应日益复杂的自主AI系统
• 劳动力断层：企业引入AI工具的速度是员工培训速度的2-3倍
• 算力军备赛：英伟达联手英国初创公司Ineffable Intelligence，打造下一代AI训练基础设施
• 中小企业破局：Anthropic推出新版Claude，降低AI使用门槛
• 万亿级商机：美国自主电商代理收入2030年预计达1万亿美元
• 私有模型升温：企业为掌控数据与安全，纷纷转向私有化部署方案
• 机器人触觉突破：博世与研究人员开发“触觉梦境”AI系统，提升人形机器人精细操作能力

中文翻译：

由谷歌云赞助
选择你的首个生成式AI应用场景
要开始使用生成式AI，首先聚焦于那些能够改善人类与信息交互体验的领域。

随着AI技术日趋成熟，企业正努力确保日益自主化和互联化的系统安全。

编者按：欢迎来到《提示》栏目，这是为您提供的每周AI领域动态简报。我们以分析视角解读本周重大进展，并附上真正值得关注的故事精选。

AI正同时成为网络安全工具与网络安全威胁。
那些旨在提升效率、实现工作流自动化的AI系统，同样在创造新的攻击面、治理挑战与运营风险。
本周的报道清楚表明，行业已进入AI安全防护的新阶段。挑战不再局限于保护模型或数据，企业如今正试图确保那些能够采取行动、跨工作流交互并在企业环境中日益自主运行的AI系统的安全。
而企业也逐渐意识到，传统安全方法已无法满足运营级AI系统的需求。

本周，OpenAI推出了Daybreak网络安全计划，这反映出更广泛的行业趋势：将韧性构建到AI系统中，而非仅依赖被动防御。
随着企业越来越多地采用AI系统——其速度往往快于对潜在风险的认知——Daybreak提供了漏洞防护。
OpenAI的Daybreak发布也表明，AI网络安全正日益成为生态系统的协同努力，包括思科、CrowdStrike和Cloudflare在内的企业都参与了该计划。

与此同时，AI防御与AI风险之间的界限正变得模糊。
随着AI系统日益自主化并嵌入企业运营，它们也变得更难监控、治理和保护。挑战已不再只是保护模型或数据，企业越来越需要管理那些能够在不同工作流和环境中以更高自主性运行的互联系统。
近期涉及教育科技平台Canvas的安全漏洞提醒我们，许多组织在将更多AI融入现有系统之前，已在努力管理日益互联的系统。

挑战已超越构建AI系统本身，而是要在大规模企业环境中安全可靠地将其投入运营。
谷歌云推动招聘AI部署工程师，同时OpenAI推出独立的咨询业务，这凸显了企业级AI部署在实践中仍有多困难。组织采用AI的速度快于培训员工使用它的速度，而可观测性、治理和基础设施系统则难以跟上日益智能体化、互联化的环境节奏。
这正将焦点从单一的模型转向大规模部署、管理和保护AI所需的运营系统。挑战不再是构建AI系统，而是当它们日益自主化并嵌入运营时，如何安全地部署和管理它们。

除了网络安全和治理，本周的报道还强调了AI采用如何重塑企业运营、劳动力准备程度以及下一代基础设施。
为何AI正迫使企业重新思考可观测性：AI系统正变得日益复杂和自主，迫使企业重新思考传统的可观测性工具和监控策略。
雇主采用AI工具的速度快于培训员工使用它们：许多组织采用AI工具的速度快于培训员工有效使用它们的能力，这造成了劳动力准备度和生产率的新缺口。
英伟达牵手英国AI初创公司构建“AI下一个前沿”：英伟达正与英国初创公司Ineffable Intelligence合作，帮助构建下一代AI训练基础设施，凸显出对算力和模型开发能力的持续需求。
Anthropic瞄准小企业推出最新Claude版本：Anthropic正通过发布新版本Claude，拓展对小企业市场的布局，旨在让生成式AI更易被小型组织使用。
美国代理商务收入预计到2030年将达到1万亿美元：到2030年，美国代理商务收入预计将达到1万亿美元，这标志着人们对能够代表消费者进行购物、推荐和完成交易的AI系统的信心日益增强。
企业为何转向私有AI模型：企业正越来越多地探索私有AI模型，以期对数据、安全以及AI系统在企业内的部署方式获得更大控制权。
博世携手研究人员开发人形机器人灵巧手AI：博世开发了一种名为“触觉梦境”的新型AI驱动系统，用于提升人形机器人的灵活性与真实环境下的表现。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
As AI technology becomes more established, enterprises are struggling to secure increasingly autonomous and interconnected systems.
Editor’s Note: Welcome to Prompt, your weekly briefing on the shifting AI landscape. We provide an analytical look at the week’s biggest developments, paired with a curated roundup of the stories that actually matter.
AI is becoming both a cybersecurity tool and a cybersecurity threat.
The same AI systems designed to improve efficiency and automate workflows are creating new attack surfaces, governance challenges and operational risks.
That’s evident in this week’s coverage, which shows the industry has entered a new phase of securing AI. The challenge is no longer just protecting models or data. Companies are now trying to secure AI systems that can take action, interact across workflows and increasingly operate on their own inside enterprise environments.
And companies are realizing that traditional security approaches aren’t enough for operational AI systems.
This week, OpenAI launched Daybreak, a cybersecurity initiative that reflects a broader shift toward building resilience into AI systems rather than relying solely on reactive defenses.
It offers vulnerability protection as companies increasingly adopt AI systems, often faster than they can understand potential risks.
OpenAI’s Daybreak rollout also points to how AI cybersecurity is increasingly becoming an ecosystem effort, with companies including Cisco, CrowdStrike and Cloudflare participating in the initiative.
Meanwhile, the line between AI defense and AI risk is blurring.
As AI systems become more autonomous and embedded in enterprise operations, they also become harder to monitor, govern and secure. The challenge is no longer just protecting models or data. Companies are more and more trying to manage interconnected systems that can act across workflows and environments with greater autonomy.
Recent breaches involving the ed tech platform Canvas are a reminder that many organizations are already struggling to manage increasingly connected systems before adding more AI into the mix.
The challenge has moved beyond building AI systems. It’s operationalizing them safely and reliably at enterprise scale.
Google Cloud’s push to hire AI deployment engineers, alongside OpenAI’s launch of a standalone consulting business, highlights how difficult enterprise AI deployment remains in practice. Organizations are adopting AI faster than they can train employees to use it, while observability, governance and infrastructure systems struggle to keep pace with increasingly agent-rich interconnected environments.
That’s shifting the focus away from models alone and toward the operational systems required to deploy, manage and secure AI at scale.
The challenge is no longer just building AI systems. It’s figuring out how to securely deploy and manage them at enterprise scale as they become more autonomous and embedded into operations
Beyond cybersecurity and governance, coverage also highlighted how AI adoption is reshaping enterprise operations, workforce readiness and the next generation of infrastructure.
Why AI Is Forcing Enterprises to Rethink Observability: AI systems are becoming more complex and autonomous, forcing enterprises to rethink traditional observability tools and monitoring strategies.
Employers Take On AI Tools Faster Than They Can Train Workers to Use Them: Many organizations are adopting AI tools faster than they can train employees to use them effectively, creating new gaps in workforce readiness and productivity.
Nvidia Taps British AI Startup to Build ‘Next Frontier’ of AI: Nvidia is partnering with British startup Ineffable Intelligence to help build next-generation AI training infrastructure, underscoring continued demand for compute and model development capacity.
Anthropic Targets Small Businesses With Latest Claude Release: Anthropic is expanding its push into the small business market with a new Claude version release designed to make generative AI more accessible to smaller organizations.
US Agentic Commerce Revenue Forecast to Reach $1 Trillion by 2030: Agentic commerce revenue in the U.S. is projected to reach $1 trillion by 2030, signaling growing confidence in AI systems that can shop, recommend and complete transactions on behalf of consumers.
Why Companies Are Shifting Toward Private AI Models: Companies are increasingly exploring private AI models to gain greater control over data, security, and how AI systems are deployed within the enterprise.
Bosch, Researchers Develop AI for Humanoid Dexterity: Bosch has developed a new AI-driven “touch dreaming” system to improve the dexterity and real-world performance of humanoid robots.