字号·· | 护眼
商业内幕

老板们正重新考虑员工能否使用某些AI工具Bosses are rethinking whether you get access to certain AI tools

点「原文对照」整页切到原文,或双击某段只看那段的原文。

你最喜欢的工作用AI工具很快可能会配备预算、把关人,或者两者兼有。在多年敦促员工拥抱这项技术之后,包括高盛、思科和摩根大通在内的公司正在采取更具针对性的方式部署AI。从设定预算、追踪使用情况,到限制特定员工或工具的访问权限,这些不同的策略出台之际,各公司正在向AI投入越来越多的资金,并试图降低安全风险。

Your favorite AI tool at work might soon come with a budget, a gatekeeper, or both. After years of urging staff to embrace the technology, companies including Goldman Sachs, Cisco, and JPMorgan are taking a more targeted approach to deploying it. From setting budgets and tracking usage to restricting access to certain employees or tools, the different strategies come as companies pour ever-larger sums into AI and seek to reduce security risks.

许多雇主正在意识到,AI使用“正在迅速普及”,他们需要设置护栏,麦肯锡公司科技、媒体和电信业务高级合作伙伴Lari Hämäläinen表示。在高盛,首席执行官兼董事长大卫·所罗门本月表示,尽管该银行正在迅速扩大向其4.5万名员工提供的AI工具范围,但让所有员工每天都能使用最先进的模型并不合理。“对于某些事情,特别是网络安全,我们需要前沿模型,”他对彭博电视表示。

Many employers are realizing that AI use is "really picking up" and they need to set guardrails, said Lari Hämäläinen, a senior partner in McKinsey & Company's technology, media, and telecommunications practice. At Goldman Sachs, CEO and chairman David Solomon said this month that while the bank is rapidly expanding the range of AI tools it provides to its 45,000 employees, it doesn't make sense for all of them to have access to the most advanced models every day. "We need frontier models for certain things, particularly cyber," he told Bloomberg TV.

但是,“有很多事情,我们不需要前沿模型”来帮助员工提高生产力。其他公司在先前采用较为宽松的政策推动采用之后,正在限制AI支出。思科首席产品官Jeetu Patel将公司早期的做法比作学习骑自行车:编码团队在担心效率之前,先获得了数月的无限制代币来熟练掌握AI,他上个月在CXOTalk播客上表示。随着使用量的增长,Patel表示,他3.2万人团队中的经理们将获得代币预算。

However, "there are a lot of things where we don't need the frontier" to help employees be more productive. Other companies are limiting AI spending after first using looser policies to drive adoption. Cisco's chief product officer, Jeetu Patel, compared the company's early approach to learning to ride a bicycle: Coding teams got months of unlimited tokens to become fluent with AI before worrying about efficiency, he said last month on the CXOTalk podcast. As usage has grown, Patel said managers across his 32,000-person team would receive token budgets.

9月,摩根大通对部分工程师使用Anthropic的Claude Code设定了每月2000美元的上限,商业内幕此前报道。一位发言人表示,该银行正在将成本与生产力、质量、风险降低和业务影响挂钩。工作场所对AI实施更严格监管的趋势,出现在一些雇主曾夸耀其员工消耗了多少代币的几个月后。

In September, JPMorgan imposed a monthly $2,000 cap on some engineers' use of Anthropic's Claude Code, Business Insider previously reported. A spokesperson said the bank was tying costs to productivity, quality, risk reduction, and business impact. The move toward tighter oversight of AI in the workplace comes months after some employers boasted how many tokens their workforces were gobbling up.

Visa表示,其每月代币使用量从2月的1万亿增至3月的近2万亿,并正在向其AI高级用户发放奖励。3月,英伟达首席执行官黄仁勋表示,该公司正试图每年为其工程师花费高达20亿美元购买代币,并称,如果一名年薪50万美元的工程师使用的代币价值不足25万美元,他将“深感震惊”。雇主们还推出了仪表盘,以使员工的AI使用更加可见——在某些情况下,还旨在鼓励更广泛的采用。

Visa said it doubled its monthly token use from 1 trillion in February to nearly 2 trillion in March and that it was giving prizes to its AI power users. In March, Nvidia CEO Jensen Huang said the company was trying to spend as much as $2 billion annually on tokens for its engineers — and that he would be "deeply alarmed" if a $500,000 engineer used less than $250,000 worth. Employers also rolled out dashboards to make workers' AI use more visible — and, in some cases, to encourage broader adoption.

例如,据《商业内幕》4月报道,迪士尼的仪表盘显示,其最大的AI用户平均每天发出数百次请求,并使用数千万个代币。同样,毕马威美国公司5月表示,希望为其咨询业务推出一个仪表盘,以鼓励该部门约1万名员工进行更“频繁和复杂”的使用。一位发言人告诉《商业内幕》,由于员工现在已在日常工作中使用AI,该公司已停止通过仪表盘监控AI使用情况。

For example, Disney's dashboard showed its biggest AI users were making hundreds of requests per day on average and using tens of millions of tokens, Business Insider reported in April. Similarly, KPMG US said in May it hoped a dashboard for its advisory business would encourage more "frequent and sophisticated" use among the unit's roughly 10,000 workers. It has since discontinued monitoring AI usage through the dashboard now that staffers are using AI in their everyday work, a spokesperson told Business Insider.

现在,不断上升的AI支出正促使公司将重点从扩大使用转向最大化投资。梅洛风险投资公司的一项新研究发现,企业AI部署支出到9月已达到1770亿美元,而2025年全年为330亿美元。与此同时,在盖特纳5月至7月对约2000名科技高管的调查中,只有13%表示其组织从AI中获得了显著价值。千年联盟联合创始人亚历克斯·索博尔表示,公司的AI账单可能在部分原因上膨胀,因为员工缺乏限制使用的动力。

Now, rising AI spending is prompting companies to shift their focus from expanding usage to maximizing their investment. A new study from Menlo Ventures found that enterprise AI deployment spending reached $177 billion by September, up from $33 billion in all of 2025. Meanwhile, only 13% of about 2,000 tech executives polled by Gartner from May through July said their organizations had gained significant value from AI. Companies' AI bills may be ballooning in part because workers have little incentive to curb their use, said Alex Sobol, cofounder of the Millennium Alliance, an invite-only community for C-suite executives.

千年联盟是一个仅限受邀参加的高管社区。他说:“没有限制,他们会以各种可能的方式使用它。”广泛的AI访问如果使用不当,也可能带来安全风险——这是公司可能变得更加谨慎的另一个原因。

Without limits, "they're going to use it in every possible way they can," he said. Broad AI access can also create security risks if used improperly — another reason companies might be getting more judicious. A 2025 KPMG survey of more than 1,000 US workers found that 44% were using AI at work without authorization or in inappropriate ways, while 46% had uploaded sensitive company information or intellectual property to public AI platforms.

一项2025年毕马威对超过1000名美国员工的调查发现,44%的人在工作中未经授权或以不当方式使用AI,而46%的人曾将敏感公司信息或知识产权上传至公共AI平台。员工管理软件初创公司Rippling在8月对AI工具设定了月度上限,此前该公司预计代币支出可能达到其研发人员预算的40%。该公司表示,这些限制以及更高效的模型路由,将该预测降至10%至15%之间,而代币使用量仍接近之前的峰值。

Rippling, a workforce-management software startup, set monthly caps in August on AI tools after projecting token spending could reach 40% of its R&D head count budget. The company said those limits and more efficient model routing lowered that projection to between 10% and 15% while token use remained near its previous peak. "The point isn't to cut AI costs," Matt MacInnis, Rippling's chief product officer, said in a statement to Business Insider.

“重点不在于削减AI成本,”Rippling首席产品官马特·麦克因尼斯在向《商业内幕》发表的声明中表示,“我们其实可以接受成本增长,前提是这种增长是有成效的。”

"We're actually fine with cost growth, provided it's productive."

阅读媒体原文 ↗