不能再随便用:高盛、思科等企业开始对员工 AI 使用设限Goldman Sachs, Cisco and Others Begin Restricting Employee AI Use


AI free token/AI免费积分

IT之家 11 月 11 日消息,据《商业内幕》报道,在过去几年不断鼓励员工拥抱 AI 工艺之后,涵盖高盛、思科和摩根大通在内的企业,正开始采取更有针对性的 AI 部署策略。从设定使用预算、追踪调用量,到限制特定员工或工具的访问权限,这些不同的管理措施背后,是企业在不断加大 AI 投入的同时,也希望降低相关安全风险。

不能再随便用:高盛、思科等企业开始对员工 AI 使用设限

麦肯锡公司(McKinsey & Company)工艺、媒体和电信业务高级合伙人拉里 · 哈马莱宁(Lari Hämäläinen)表示,越来越多的企业意识到,员工使用 AI 的频率正在“大幅上升”,因而有必要为其设定明确的使用规范和限制。

高盛董事长兼首席执行官戴维 · 所罗门(David Solomon)本月表示,尽管该银行正在迅速扩大 AI 工具的覆盖范围,让旗下 4.5 万名员工都能使用更多 AI 机型,但并没有必要让所有员工每天都能访问最先进的模型。

他在接受彭博电视采访时表示:“某些工作确实需要最前沿的模型,尤其是网络安全领域。”但他同时指出,“很多工作并不需要最前沿的模型”,普通 AI 工具就足以协助员工提高质率。

另一些企业则在最初采取相对宽松的政策推动员工使用 AI 后,开始限制相关开支。

思科首席机型官吉图 · 帕特尔(Jeetu Patel)上个月在 CXOTalk 播客节目中,将公司早期的 AI 推广策略比作学习骑自行车。当时,思科允许开发团队连续数月无限量使用 AI 编程工具的 Token,让员工先熟悉并掌握 AI 工艺,而不是一开始就考虑使用效率。

随着使用量不断增长,帕特尔表示,他计划为旗下具备 3.2 万名员工的团队中的各级管理人员设定 Token 使用预算。

此前,《商业内幕》报道称,摩根大通已于今年 9 月对部分工程师使用 Anthropic 旗下 Claude Code 的额度设置了限制,每月上限为 2,000 美元(IT之家注:现汇率约合 13,376 元人民币)。

该银行发言人表示,公司正在将 AI 使用成本与生产效率、工作质量、风险降低程度以及实际业务影响挂钩。

企业开始加强对员工 AI 使用情况的管理,距离一些公司公开炫耀员工消耗了多少 Token,实际上还不到几个月。

Visa 曾表示,公司每月的 Token 使用量从今年 2 月的 1 万亿增长至 3 月的近 2 万亿,并为 AI 使用量最大的员工给出奖励。

今年 3 月,英伟达首席执行官黄仁勋(Jensen Huang)表示,公司计划每年为工程师使用 Token 投入最高 20 亿美元(现汇率约合 133.76 亿元人民币)。他甚至开玩笑称,如果一名年薪 50 万美元(现汇率约合 334.4 万元人民币)的工程师只使用了价值不到 25 万美元(现汇率约合 167.2 万元人民币)的 Token,他会“感到格外担忧”。

与此同时,一些企业还推出了数据看板,让管理层可以更直观地了解员工使用 AI 的情况,有时也借此鼓励更多员工参与。

例如,今年 4 月迪士尼推出的 AI 使用数据看板表明,公司使用 AI 最频繁的员工平均每天会发起数百次请求,Token 消耗量达到数千万级别。

类似地,毕马威美国(KPMG US)今年 5 月表示,希望借助为咨询业务部门推出数据看板,鼓励旗下约 1 万名员工更加频繁、更加深入地使用 AI。

不过,随着员工逐步将 AI 融入日常工作,毕马威当下已经停止借助该数据看板监控 AI 使用情况。公司发言人透露,相关工具已经完成了原本的推广任务。

企业开始追求每个 Token 的实际价值

如今,AI 开兼容续攀升,促使企业将管理重点从扩大使用规模转向提高投资回报。

风险投资机构 Menlo Ventures 的一项最新研究表明,截至今年 9 月,企业在 AI 部署维度的支出已达到 1,770 亿美元(现汇率约合 1.18 万亿元人民币),而 2025 年全年这一数字仅为 330 亿美元(现汇率约合 2,206.96 亿元人民币)。

但投入增长并不意味着企业已经取得相应回报。Gartner 在今年 5 月至 7 月对约 2000 名科技行业高管开展的一项调查表明,只有 13% 的受访者表示,其所在企业已经从 AI 工艺中取得了显著价值。

Millennium Alliance 联合创始人亚历克斯 · 索博尔(Alex Sobol)表示,企业 AI 账单不断膨胀,部分原因在于员工缺乏控制使用量的动力。

Millennium Alliance 是一个面向企业高管、采取邀请制的交流组织。索博尔指出,如果企业不设定使用限制,员工“就会想尽一切办法使用 AI”。

另外,如果使用不当,向员工开放广泛的 AI 访问权限还可能带来安全隐患,这也是企业开始更加谨慎管理 AI 使用情况的另一个原因。

毕马威 2025 年针对 1000 多名美国员工开展的一项调查表明,44% 的受访者曾在未经授权或不恰当的情况下,在工作中使用 AI;另有 46% 的受访者曾将公司敏感信息或知识产权上传至公开 AI 平台。

人力资源管理软件初创公司 Rippling 也采取了类似措施。今年 8 月,公司开始对 AI 工具设置月度使用上限。此前,公司预计 Token 支出可能达到研发团队总体人力成本预算的 40%。

Rippling 表示,借助实施使用限制并优化模型调用路由,Token 支出占比的预测值已经降至 10% 至 15%,而 Token 使用量仍基本维持在此前的峰值水平。

Rippling 首席机型官马特 · 麦金尼斯(Matt MacInnis)强调:“笔者的目的不是削减 AI 成本。只要这些支出可以带来实际生产价值,笔者完全可以接受成本继续增长。”

本文来源:https://www.ithome.com/1/025/734.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT Home, November 11 — According to Business Insider, after years of encouraging employees to embrace AI, companies including Goldman Sachs, Cisco, and JPMorgan Chase are now adopting more targeted AI deployment strategies. From setting usage budgets and tracking token consumption to limiting access for specific employees or tools, these management measures reflect companies' desire to reduce security risks while continuing to increase their AI investments.

Goldman Sachs, Cisco and Others Begin Restricting Employee AI Use

Lari Hämäläinen, senior partner in McKinsey & Company's technology, media, and telecommunications practice, said that more and more companies realize that employees' AI usage frequency is "rising significantly," making it necessary to set clear usage guidelines and restrictions.

Goldman Sachs Chairman and CEO David Solomon said this month that although the bank is rapidly expanding AI tool coverage to give its 45,000 employees access to more AI models, there is no need for all employees to access the most advanced models daily.

In an interview with Bloomberg Television, he said: "Some jobs really need the most cutting-edge models, especially in cybersecurity." But he also noted that "many jobs don't need the most cutting-edge models," and ordinary AI tools are sufficient to help employees improve quality and efficiency.

Other companies, after initially adopting relatively lenient policies to promote employee AI use, are now beginning to limit related spending.

Cisco Chief Model Officer Jeetu Patel, speaking on the CXOTalk podcast last month, compared the company's early AI promotion strategy to learning to ride a bicycle. At that time, Cisco allowed development teams to use AI coding tool tokens without limits for several consecutive months, letting employees first become familiar with and master AI technology, rather than considering usage efficiency from the start.

As usage continues to grow, Patel said he plans to set token usage budgets for managers at all levels within his team of 32,000 employees.

Previously, Business Insider reported that JPMorgan Chase had set limits on some engineers' use of Anthropic's Claude Code in September this year, with a monthly cap of $2,000 (IT Home note: approximately 13,376 RMB at current exchange rates).

A bank spokesperson said the company is linking AI usage costs to productivity, work quality, risk reduction, and actual business impact.

Companies are beginning to tighten management of employee AI usage, less than a few months after some companies publicly boasted about how many tokens their employees were consuming.

Visa once stated that the company's monthly token usage grew from 1 trillion in February this year to nearly 2 trillion in March, and it gave rewards to employees who used the most AI.

In March this year, NVIDIA CEO Jensen Huang said the company plans to invest up to $2 billion annually for engineers' token usage (approximately 13.376 billion RMB at current exchange rates). He even joked that if an engineer with a $500,000 annual salary (approximately 3.344 million RMB at current exchange rates) only used less than $250,000 (approximately 1.672 million RMB at current exchange rates) worth of tokens, he would be "particularly worried."

Meanwhile, some companies have also launched data dashboards to give management a more intuitive view of employee AI usage, sometimes also using them to encourage more employees to participate.

For example, Disney's AI usage data dashboard launched in April this year showed that the company's employees who use AI most frequently initiate hundreds of requests on average per day, with token consumption reaching tens of millions.

Similarly, KPMG US said in May this year that it hoped to encourage approximately 10,000 employees to use AI more frequently and deeply by launching a data dashboard for its consulting business department.

However, as employees gradually integrate AI into daily work, KPMG has now stopped using the data dashboard to monitor AI usage. A company spokesperson revealed that the relevant tool had completed its original promotional task.

Companies begin to pursue the actual value of each token

Now, as AI costs continue to climb, companies are shifting their management focus from expanding usage scale to improving return on investment.

A recent study by venture capital firm Menlo Ventures shows that as of September this year, companies' spending on AI deployment had reached $177 billion (approximately 1.18 trillion RMB at current exchange rates), while for the full year 2025, this figure was only $33 billion (approximately 220.696 billion RMB at current exchange rates).

But rising investment does not mean companies have already obtained corresponding returns. A survey Gartner conducted from May to July this year among about 2,000 tech-industry executives showed that only 13% of respondents said their companies had already achieved significant value from AI processes.

Millennium Alliance co-founder Alex Sobol said that corporate AI bills keep ballooning partly because employees lack incentives to control usage.

Millennium Alliance is an invitation-only networking organization for corporate executives. Sobol noted that if companies do not set usage limits, employees “will use AI by every means possible.”

In addition, granting employees broad AI access can also create security risks if used improperly, which is another reason companies have begun managing AI usage more cautiously.

A KPMG 2025 survey of more than 1,000 U.S. employees found that 44% of respondents had used AI at work without authorization or in an inappropriate manner; another 46% had uploaded sensitive company information or intellectual property to public AI platforms.

HR management software startup Rippling has taken similar steps. In August this year, the company began setting monthly usage caps for AI tools. Previously, it estimated that Token spend could reach 40% of the research and development team’s total labor cost budget.

Rippling said that by implementing usage limits and optimizing model-invocation routing, the projected share of Token spend has fallen to 10% to 15%, while Token usage has remained largely at its prior peak level.

Rippling Chief AI Officer Matt MacInnis emphasized: “My goal is not to cut AI costs. As long as these expenditures deliver real productivity value, I’m perfectly fine with costs continuing to rise.”

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry