Companies Are Losing Fortunes to Unlimited Employee AI Access

The American fintech startup Slash, valued at $1.4 billion, publicly admitted that one of its employees spent $81,267 in AI credits building a simple shooter game featuring internet memes. The company itself had encouraged the team to code intensively with AI just a week earlier.

An $80,000 Game

Nicolas Brilliante, head of strategic industries at Slash, used a company card to generate Brainrot Shooter, a game styled after Minecraft in which players shoot characters with internet-meme names like skibidi toilet. He admitted on X that it was a genuine mistake, saying he had underestimated his own capabilities.

Slash responded with a wink, suggesting it might file the expense as a marketing cost and encouraging others to play the game. Brilliante later joked that the incident might become a training case study on how AI spending can spiral out of control.

A Phenomenon With a Name

Slash's story fits into a broader trend known as tokenmaxxing, where employees maximize their AI token usage after being encouraged by their own companies to use these tools without clear limits. According to reports cited by Spider's Web, one American company ran up a $500 million bill with Anthropic due to unrestricted employee access to the Claude chatbot.

At Accenture, the problem mainly affects departments outside engineering. Justice Kwak, the company's head of AI strategy, admitted that the token usage isn't coming from developers' work, but from the fact that many non-IT employees use AI tools for simple tasks like converting PDF files into PowerPoint presentations. Some teams were reportedly told that too little AI usage could hurt their annual reviews and raises.

Companies Start Tightening the Reins

Uber went in the opposite direction and introduced hard caps, limiting AI token usage to $1,500 per employee per month. It's a sign that executives' enthusiasm for unlimited access to generative tools is giving way to cost control as the bills start to seriously strain budgets.

For Polish companies deploying Claude, ChatGPT Enterprise, or Copilot, this is a warning that mere access to a tool, without clear rules and business goals, doesn't automatically translate into productivity. A poorly thought-out rollout, based solely on the slogan more AI at work, can end up generating costs comparable to the salaries of entire teams instead of savings.

What Comes Next

The examples of Slash, Uber, and Accenture show that the market is entering a correction phase after the first wave of unreflective AI adoption in daily work. Tech companies, including Anthropic and OpenAI themselves, may begin in the coming months to offer business customers better tools for monitoring and capping token usage, rather than relying solely on employees' self-discipline.

Sources: Spider's Web (spidersweb.pl), American Bazaar Online (americanbazaaronline.com).