Major Companies Throttle Employee AI Access as Costs Exceed Budgets

Breitbart ^ | July 5, 2026 | Lucas Nolan

[chickenlips]

Technology, banking, and entertainment companies are significantly restricting employee access to AI tools as monthly costs skyrocket into tens of millions of dollars, according to internal documents and communications obtained from multiple organizations.

404 Media reports that companies across multiple industries are implementing strict limitations on employee use of AI tools after experiencing dramatic cost increases that have caught executives off guard. Internal communications from organizations including Atlassian, Adobe, Amazon, Citi, and GitHub reveal a growing crisis as AI spending triples in some cases, forcing companies to cut access to powerful models and plead with workers to reduce consumption.

The situation stems from two converging factors: companies rapidly adopting AI without fully understanding the cost implications, and AI providers transitioning from flat-rate subscription models to usage-based billing that charges based on the number of AI tokens consumed during each interaction.

Citi has taken particularly aggressive action, completely shutting off employee access to the latest versions of Claude and ChatGPT models. According to an internal email, the bank disabled Claude Opus 4.6 and 4.7, as well as GPT-5.5, on June 24. The email explicitly states these models “consume significantly more AI Credits per interaction and have been the primary driver of elevated enterprise consumption.”

Prior to the shutdown, Citi sent employees warnings urging them to choose less powerful models unless absolutely necessary. The communication explained that since AI tokens are pooled across the entire organization, employees with heavier AI-assisted workflows draw more from shared resources. The email provided specific guidance on which models to use for different tasks, recommending GPT-5.3-Codex for simple questions and code generation, while reserving higher-tier models like Claude Sonnet 4.6 for architectural reasoning.

Citi’s changes came in direct response to a shift from flat subscription pricing to usage-based billing in June.

---

TOPICS: Business/Economy
KEYWORDS: ai; business; expense; panicporn

[chickenlips]

Just shows that full automation of the surveillance state is going to be costly.

[Red Badger]

What are they charging by the second or by the byte?...........

[catnipman]

i wonder much of that expensive AI power is being used to generate high-quality porn videos, naturally tailored to the specific imaginations of each employee?

[T.B. Yoits]

"AI" isn't artificial and it is not intelligent.

These korporations really thought they were going to outsource this work to Indian programmers, while calling it "AI".

If these "programmers" were any good at it, India wouldn't look like India.

AI startup was 700 Indian programmers "AI Chatbot Turns Out to Be 700 Engineers in India":
https://tech.co/news/ai-startup-chatbot-revealed-as-human-engineers

'Amazon Abandons Checkout-Less Tech That Secretly Used 1,000 Indian Employees Watching Cameras'
https://thedeepdive.ca/amazon-abandons-checkout-less-tech-that-secretly-used-1000-indian-employees-watching-cameras/ "AI" isn't artificial and it isn't intelligent.

"Amazon is doing away with the “Just Walk Out” checkout-less technology at its Fresh grocery stores... Instead of relying solely on artificial intelligence and advanced sensors, as Amazon had claimed, the Just Walk Out system depended heavily on manual labor from over 1,000 workers based in India... These remote “cashiers” monitored video footage from stores to track what items customers removed from shelves before generating receipts hours later."

[PAR35]

I know of one company that charges by the query for database search work (not generative work). Want to revise your query? that’s another hefty charge.

For those of us who were raised on using a series of queries to narrow or broaden a search until you find what you need, that’s a no-go.

[Red Badger]

These companies should develop their own in-house AI systems...........

[ClearCase_guy]

Corporate Guy #1: “We are going to have to lay off a bunch of you mooks. AI is cheaper. Sorry about that.”

Corporate Guy #2: “And to those of you who are still with us, please limit your use of AI. It’s too expensive.”

Corporate Guy #3: “Good news boss — profits are way up due to all of our cost-cutting efforts.”

[E. Pluribus Unum]

I am running qwen3-coder-next:80b on my local Linux Mint Ollama server with 64GB system RAM and 16GB VRAM that I purchased for $1200 and spent another $150 for 32GB more RAM.

It is almost as good as Claude.ai for coding and far cheaper.

Double the VRAM and system RAM and it could handle 6-12 concurrent clients.

Local AI is the way to go.

[PAR35]

That sounds like those ‘self driving’ cars that used folks in the Philippines to guide the vehicles.

[House Atreides]

“ That sounds like those ‘self driving’ cars that used folks in the Philippines to guide the vehicles.”
************************************

Look… pay no attention to that man behind the curtains!

[Right_Wing_Madman]

"AI" isn't artificial and it is not intelligent. These korporations really thought they were going to outsource this work to Indian programmers, while calling it "AI".

[E. Pluribus Unum]

What are they charging by the second or by the byte?...........

By the token. https://x.com/i/grok/share/5af5c3dd92de449284a05e1f66680836

[ProtectOurFreedom]

Unsaid is the huge fear that AI will exfiltrate IP, financial data, employee records, trade secrets, and other confidential information. That might actually be a far bigger reason for the cut-back and they are just using cost as the public-facing excuse.

[ProtectOurFreedom]

Corporate Guy #4 tomorrow: “Bad news boss — customers are leaving in droves and the top line is collapsing. We can’t support them.”

[Fury]

Good post. One of our clients has a restrictive AI policy for this reason.

[fso301]

i wonder much of that expensive AI power is being used to generate high-quality porn videos, naturally tailored to the specific imaginations of each employee?

Probably a lot more than most realize. Porn was a driver in the internet boom internet 30 years.

[FrankRizzo890]

They charge by “token”, which you can’t seem to get a good definition of.

[Red Badger]

What the heck is a ‘token’?............

[FrankRizzo890]

Back in ‘92 I designed and built the datacenter in what is now Change Healthcare (we were MediFAX back then). We *ALWAYS* charged by the transaction which usually corresponded to a database hit. (Either locally on my system, or on a system belonging to an insurance company, or a state medicaid provider).

Our stuff was to verify that a patient WAS covered by Medicaid, and what was covered, and how much. These weren’t overly expensive transactions and the hospitals LOVED them. (If they got a verification from us, the state would almost always have to pay. And I never remember hearing about a case where the state contested their previous response).

[No.6]

Yep; I’ve been touting smaller models and focused data for specific applications but … nobody want to miss being the one to make a real general purpose AI.

In the meantime the smart people will follow you. The pendulum swings back off centralization (cloud) and back to “on-prem.”