Goldman, Cisco and JPMorgan Put AI Tool Access on Budgets

By Nathan Brooks

JPMorgan capped some engineers’ Claude Code use at $2,000 a month in September, as employers moved from broad AI access to budgets, employee limits and tool controls.

Employees may now face a budget, a gatekeeper, or both.

After years of urging staff to embrace AI, Goldman Sachs, Cisco and JPMorgan now favor targeted deployment over blanket access. The stated aim is not to halt spending. It is to get more from large AI commitments while reducing security risk.

New controls now include budgets measured in AI tokens. Tokens are the units AI models use to process text and code. Usage dashboards make AI activity visible to employers. Employers also limit access by employee and by tool.

Across the market, companies still pour ever-larger sums into AI while returns remain uncertain. Lari Hämäläinen, a senior partner in McKinsey & Company’s technology, media and telecommunications practice, said AI use was picking up rapidly and employers needed guardrails.

What changes for employees?

At Goldman Sachs, CEO and chairman David Solomon said the bank was rapidly expanding AI tools for 45,000 employees, but not giving everyone daily access to frontier models (the most advanced AI systems).

This month, Solomon told Bloomberg TV: “We need frontier models for certain things, particularly cyber ... there are a lot of things where we don’t need the frontier” to help employees be more productive.

Last month on the CXOTalk podcast, Cisco chief product officer Jeetu Patel compared the early approach with learning to ride a bicycle. Coding teams had months of unlimited tokens to build fluency before efficiency became the concern. As usage grew, Patel said managers across Cisco’s 32,000-person team would receive token budgets.

The JPMorgan cap applied to some engineers using Anthropic’s Claude Code. A spokesperson said the bank tied costs to productivity, quality, risk reduction and business impact.

How did AI usage become so large?

These controls arrived after employers publicized much higher AI consumption.

Visa, the payments company, doubled its monthly token use from 1 trillion in February to nearly 2 trillion in March, while also 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 engineers. He would be deeply alarmed if a $500,000 engineer used less than $250,000 worth.

Employers rolled out dashboards to make AI use visible. Some dashboards also aimed to encourage broader adoption.

Disney’s dashboard showed its biggest AI users making hundreds of requests per day on average. They were using tens of millions of tokens, Business Insider reported in April.

KPMG US said in May that a dashboard for its advisory unit would encourage more frequent and sophisticated use. The unit had roughly 10,000 workers. KPMG later discontinued dashboard monitoring because AI had become part of everyday work, a spokesperson said.

What do the spending figures reveal?

Menlo Ventures’ new study found that enterprise AI deployment spending reached $177 billion by September. It rose from $33 billion in all of 2025.

Gartner polled about 2,000 technology executives from May through July. Only 13% said their organizations had gained significant value from AI.

The gap supports closer management of usage, although it does not establish that fewer tokens or lower spending would close the value gap on its own.

Alex Sobol, cofounder of the invite-only Millennium Alliance for C-suite executives (senior leaders), said workers would use AI in every possible way without limits. He said employees had little incentive to curb usage on their own. Those incentives may help push corporate AI bills higher.

Can tighter controls pay for themselves?

A 2025 KPMG survey of more than 1,000 US workers found 44% were using AI at work without authorization or inappropriately. It also found 46% had uploaded sensitive company information or intellectual property to public AI platforms.

Rippling set monthly caps on AI tools in August after projecting token spending could reach 40% of its research and development headcount budget.

Those limits and more efficient model routing (assigning work to suitable models) lowered the projection to between 10% and 15%. Token use remained near its previous peak.

Rippling chief product officer Matt MacInnis said: “The point isn’t to cut AI costs ... We’re actually fine with cost growth, provided it’s productive.”

None of the examples reports what happened to output or security after controls took effect. The evidence shows that employers can meter AI. It does not show that every cap will improve productivity or security. Leadership teams should demand productivity, quality and risk measures before treating access limits as a return on investment.