Factually Impossible: Axios, the Unnamed AI Consultant and $500M Overspend.

The Flashing Red Siren of Fiction

This story captured a lot of attention and an astonishing amount of retweets.

Joyous vindication for the AI sceptics as Anthropic is shown to be desperate and even fraudulent.

“An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees.”

Why This Is Mathematically Impossible

• For Anthropic, this $500M would be recorded as $6B in ARR terms.

• Assuming Anthropic’s ARR for April was about $42B:

  • 14% of Anthropic’s revenue and 50% of their M-o-M growth came from a billing error by a customer.

  • The customer unwittingly had become the largest AI token spender in the world.

• Yet Anthropic did not mention this at all. They were extending $500M of credit to a company without telling them.

Why This Is Regulatorily Impossible

• A $500M surprise adverse event in a single month would trigger an 8-K filing with the SEC, highlighting to investors this event of a catastrophic surprise loss.

• If the US still had any regulators, buses of them would be descending on Anthropic and this customer to understand what happened. The Fed, the SEC and the DoJ would be the first of many waves.

Why This Is Organisationally Impossible

• This could only be a post-paid API bill agreement if as described.

• Implicitly without any prior flag to the customer, Anthropic presented $500M for the previous month. The article does not say whether Anthropic told them they were the world’s number 1 AI spender.

• As the bill was paid, the customer made no protest, accepting their own folly for omitting limits

• The AI consultant sounds like a poor one if their expertise did not include enterprise-destroying protections in spend.

• The CFO, the VP of Finance. Many people in that chain would, of course, have to confess to the CEO how this happened, and the CEO would understand that accidents happen, and nobody has been apparently dismissed.

Why the Source Is Impossible

• This is a nuclear secret to this company as they work out how to deal with it. The only people who would know would be:

  • A half dozen of the finance function

  • The CEO

  • The auditors

• Conceivably an MBB-style consultant working with the board.

• Now a fleet of PR / Crisis consultants would have joined.

• The way the quote is written, it sounds like a contractor happened to observe this at one of their many clients.

• The AI consultant who failed to put the use limit on board. Sounds like if anyone was to blame, it would be them, yet they chose to tip off not the FT, nor Bloomberg, they chose Axios.

Why This Is Impossible to Be Run As-is By a Credible Publisher

• The scale of this number in relation to Anthropic revenue, would make this possibly an FTX-level headline event

• The editorial process would have been exhaustive.

  • Source validated.

  • Second source required for publication backing.

  • Legal review.

  • Full editorial review.

  • Right of reply to Anthropic

  • Right of reply to the reckless yet promptly paying customer

• Then swamp the front page and channels.

Why Axios Highlight in the Article they are Not

• The revelation was a bullet point buried, without further commentary of its astonishing implications on corporate governance. Forming a pattern with other analytical conclusions:

• “One CTO told Axios that employees were using AI models to check the weather. That gets expensive fast:”

  • Top models would incur not closet to $0.005 for that. Would get expensive slow.

• Most people default to automating tasks they dislike rather than tasks most valuable to the company,” Sophia Velastegui, CEO of Velastegui Ventures and former chief AI officer at Microsoft”

  • An unsupported truism from a ‘CEO’ of a single person firm, who also never had the prior job cited.

• “Uber’s COO said AI costs are getting “harder to justify.”

  • A total mis-representation of a quote about attribution. No possible honest reading of that interview could conclude the CEO stating the justification was easier before but market dynamics were making it harder.

The Wire (Season 5)

• IYKYK

• This article portrays the values of the publication, which by this read, are atotal disregard for accurately interpreting facts.

• Unsourced quotations displaying zero industry knowledge, frustratingly veiling from whom the idiocy emanated from.

• The coup de grace, however, stands alone:

• Without comment or highlight Madison Mills exposed that a single customer contributed a 7th of Anthropic’s ARR due to a fat-fingered contract, which the customer paid, implying they absorbed the shock. Experience can be an expensive education.

I earnestly hope somebody with a breadth of reach, and heft of legitimacy, takes up this mantel.

A sword of truth needs to hack at these destructive and deceptive piles of slop.