Companies in tech, finance, and media are restricting employee access to advanced artificial intelligence models because the cost of usage has become unmanageable. Internal communications obtained by 404 Media reveal that organisations including Atlassian, Adobe, and Amazon are instructing staff to abandon high-end tools and switch to cheaper alternatives. In at least one instance, monthly AI spending has tripled to over $15 million.
Cost controls and model restrictions
The shift follows a change in how AI providers charge enterprises, moving from flat fees to usage-based billing. This means every token consumed adds directly to the company bill. Emails show some firms cutting access to specific models entirely to stop burning through their budgets. Adobe is ending unlimited access to Claude for its workforce.
An Adobe employee told 404 Media that teams were asked to adjust workflows using lower-reasoning models to reduce token consumption. The staff member noted that not everyone had fully absorbed the news or understood the full impact of the changes.
Citi has shut off access to the latest versions of Claude and ChatGPT. This includes Claude Opus 4.6 and 4.7, as well as GPT-5.5. An internal email states these models consume significantly more credits per interaction and were the primary driver of elevated enterprise consumption. The bank disabled access on June 24 and plans to re-enable them on July 1.
Before the shutdown, Citi asked staff not to use the powerful models unless absolutely necessary. One directive instructed employees to choose the right model for the task to reduce Opus 4.7 usage. Since AI tokens are pooled across the firm, heavy users draw from shared resources while lighter users ideally free up unused portions. The email stated that everyone needs to be intentional about model selection to ensure fair access.
The guidance breaks down model usage as follows:
- GPT-5.3-Codex for quick questions, explanations, or simple code generation.
- Claude Sonnet 4.6 for code review and standard chat.
- Higher models like Claude Sonnet 4.6 for architectural reasoning.
Citi confirmed it is monitoring daily Copilot usage to find excessive patterns early. The company has budget controls in place. A spokesperson told 404 Media it has not disabled models and is not allocating a fixed number of tokens to workers, despite screenshots showing access blocks.
Spending spikes at Atlassian
Atlassian ended unlimited use of its internal AI tools and introduced a dashboard where employees can track costs. The dashboard shows spending on AWS, Google Cloud, and OpenAI LLMs rose from $5 million in August 2025 to more than $15 million in May 2026. The company is on track to spend more than $120 million on AI tools for the fiscal year.
An Atlassian employee told 404 Media that staff complained about changing workflows to maximise usage, only to run out of tokens in two or three days. The employee described seeing angsty messages in Slack about how to do their jobs. The staff member said it was insane that the company allowed huge spending before, noting it was only a matter of time before it had to end.
GitHub and Amazon adjustments
At GitHub, employees do not have a limit on token spend. However, workers were told the company is looking into decreasing spend by using open source models. GitHub plans to test user-based billing, budgeting tool use to individual people instead of teams or unlimited usage.
At Adobe, unlimited Claude access is not being renewed and will expire on June 30. Staff were told to try to get everything done before that date.
Amazon recently shut down an internal leaderboard that ranked employees based on AI tool usage. Multiple staff members suspect this happened because the list encouraged wasteful and expensive usage. After the leaderboard closed, a discussion on Amazon’s internal Slack showed an employee hitting a token limit they did not know existed previously.
One Amazon employee replied on Slack: “Crazy, we go from no more leaderboard to actual usage limits in two weeks.” An Amazon spokesperson told 404 Media in an email that it encourages employees to use and experiment with AI, and guidance around usage has not changed.
Consulting firms and token economics
Other companies have hit their token limits. An employee at an entertainment company told 404 Media they hit the limit for ChatGPT token use for the first time. One developer used almost half the entire company’s allocated pool with no obvious return on investment.
Last week 404 Media reported that consulting giant Accenture found much token usage, or ‘chewing’, comes from people converting PDFs into presentation slides rather than supercharged engineers creating code. Accenture is seeing “soaring token spend” among its clients, according to leaked audio obtained by 404 Media.
There is an obvious irony in Accenture pointing this out. In the audio, senior staff explained they told clients to adopt AI as quickly as possible. Now that costs have skyrocketed, Accenture is positioning itself as the solution to that problem. One employee said the firm has a new opportunity regarding its clients: “to really think about token economics.”
Accenture continues to use AI internally for trivial projects. Screenshots obtained by 404 Media show an internal tool that lets employees predict which team will win the World Cup. The tool was made with AI.
A source with knowledge of the tool said: “They are still trying to ram AI down our throats at all levels and areas of work. Everyone seems to be trying to outdo each other in finding new ways to waste water and no one is telling us to slow down.”
Adobe, GitHub, and Accenture did not respond to requests for comment.
What it means
Workers face a new reality where using the best available tools is no longer guaranteed. Teams must actively manage their AI consumption to avoid hitting hard caps that stop work. The era of unlimited access is effectively over for many organisations.
