Rising AI expenses are forcing tech, banking, and media giants to restrict employee access to advanced models. Internal documents reveal new usage limits, model bans, and tighter monitoring across leading firms.
Escalating AI costs are prompting leading companies in technology, banking, and entertainment to impose strict limits on employee use of advanced AI models. Internal communications and dashboards from organizations such as Atlassian, Adobe, Amazon, and Citi show that executives are now restricting access to powerful AI tools and urging staff to switch to less expensive models to control spending.
According to materials reviewed by 404 Media, some companies have seen their monthly AI expenses triple, with at least one firm now spending over $15 million per month. The shift comes as AI providers move away from flat-rate pricing to usage-based billing, making every interaction with high-end models significantly more costly for enterprise customers.
At Citi, internal emails reveal that access to the latest versions of Claude and ChatGPT, including Claude Opus 4.6, 4.7, and GPT-5.5, was disabled for employees starting June 24, with plans to restore access on July 1. The company instructed staff to avoid using the most powerful models unless absolutely necessary, emphasizing that these models consume far more AI credits per interaction and are the main driver of increased costs. Citi also began monitoring daily usage of GitHub Copilot to detect excessive or unusual activity and implemented budget controls. While Citi told 404 Media it had not disabled models or capped individual usage, internal screenshots indicate otherwise.
Atlassian, known for its Jira software, recently ended unlimited AI tool access and introduced a dashboard for employees to track their AI-related spending. Data seen by 404 Media shows Atlassian's AI costs jumped from $5 million in August 2025 to over $15 million in May 2026, with projected annual spending exceeding $120 million. The company disputed the accuracy of these figures but did not clarify which numbers were incorrect. Employees reported frustration as new limits meant they could exhaust their AI allocation in just a few days, especially when using agent-based workflows or the latest Claude models.
Amazon also made changes after shutting down an internal leaderboard that ranked employees by AI tool usage. Staff speculated the leaderboard encouraged wasteful spending, and soon after, some discovered previously unknown token limits. An Amazon spokesperson said the company's guidance on AI usage had not changed and that experimentation was still encouraged.
Adobe employees were told that unlimited access to Claude would expire on June 30, with instructions to complete necessary work before the cutoff. At GitHub, there are currently no hard token limits, but employees were informed that the company is exploring ways to reduce token consumption, including testing user-based billing and considering open source models.
Other organizations are also feeling the impact. An entertainment company employee reported that their team hit its ChatGPT token limit for the first time, with one developer consuming nearly half the company's allocation without clear return on investment. Consulting firm Accenture, according to leaked audio, found that much of the AI token spend among clients comes from routine tasks like converting PDFs to slides, not just from intensive engineering work. Accenture staff described this as an opportunity to advise clients on "token economics," even as the firm continues to use AI internally for non-essential projects.
For more on how publishers are adapting to AI-driven changes, see how major media brands are redesigning their sites for agentic search.
