And what happened to them for saying it.
By Cricketpocalypse | May 2026
At some point in late 2025 or early 2026, someone at the United States Department of Defense sat down with Anthropic, one of the most valuable AI companies on earth, valued at $61 billion, and presented them with a contract.
The terms were straightforward. The Pentagon wanted to use Anthropic’s AI for government purposes. All lawful government purposes. That phrase, “all lawful purposes” is standard government contract language. Most companies sign it without blinking.
Anthropic read it. And said no.
Specifically, Anthropic refused to allow its AI to be used for two things: mass surveillance and fully autonomous weapons systems, weapons that can select and kill targets without a human making the final decision. These were Anthropic’s red lines. The things they had promised, publicly, they would not allow their technology to do.
The Pentagon blacklisted them.
WHAT HAPPENED NEXT
Within weeks, the Pentagon went to Anthropic’s competitors with the same contract and the same terms.
OpenAI said yes.
Google said yes.
In April 2026, the Pentagon reached an agreement allowing Google’s Gemini AI models to be used for — in their exact word, “any lawful government purpose.” No restrictions on surveillance. No restrictions on autonomous weapons. Any lawful purpose. Sign here.
OpenAI agreed to the same terms Anthropic had rejected.
The market rewarded yes. The market punished no.
This is not a story about evil companies doing evil things. OpenAI and Google did not cackle like villains when they signed those contracts. They did what companies in a competitive market do: they looked at the contract, they looked at the revenue, they looked at what their competitors were about to do if they didn’t sign, and they signed.
The market rewarded yes. The market punished no.
That is what makes this story more disturbing than a simple tale of greed. Nobody had to be corrupt for this outcome to happen. The system produced it automatically.
THE PROMISE THAT CAME BEFORE
To understand why this matters, you have to remember what these companies said when they were founded.
OpenAI launched in 2015 as a nonprofit. Its entire founding premise was that artificial general intelligence, AI smarter than any human, was coming, and that it was too dangerous to be controlled by any single company or government. The mission, stated publicly and repeatedly, was to develop AI “for the benefit of humanity as a whole.”
Anthropic was founded in 2021 by former OpenAI employees, including CEO Dario Amodei, who left because they believed OpenAI was moving too fast and not taking safety seriously enough. They described themselves as the responsible alternative. The company that would hold the line.
These were not marketing slogans buried in fine print. These were the founding arguments. The reasons investors gave them billions of dollars. The reasons talented researchers chose to work there instead of somewhere else. The reasons the public gave them the benefit of the doubt.
In April 2026, OpenAI president Greg Brockman sat in a federal courtroom in Oakland, California and testified under oath. He confirmed that he helped found OpenAI on the promise to advance AI for the benefit of all humanity. He confirmed that the company started as a nonprofit specifically to avoid the pressure of generating financial returns.
He also confirmed that his personal stake in OpenAI’s for-profit arm may now be worth somewhere between twenty and thirty billion dollars.
He did not explain the distance between those two things. Nobody asked him to.
THE MATH OF ETHICS IN A COMPETITIVE MARKET
Here is the problem that the Anthropic story illustrates with painful clarity.
Anthropic drew an ethical line. They said: our AI will not be used for mass surveillance or fully autonomous weapons. This was a genuine commitment with real commercial consequences, they knew the Pentagon was a massive potential customer and they drew the line anyway.
The Pentagon blacklisted them.
OpenAI and Google, watching this happen, had a choice. They could stand with Anthropic and collectively refuse to provide AI for purposes that two of the industry’s most prominent safety-focused companies had publicly called dangerous. Or they could sign the contract and take the business.
They signed the contract.
From a pure business perspective this was the rational decision. If OpenAI had refused, Google would have signed. If Google had refused, Microsoft would have signed. Someone was going to take that contract. The company that held the ethical line would lose the revenue and gain nothing, because the Pentagon would have its AI either way.
This dynamic has a name in game theory. It is called a race to the bottom. It describes situations where competitive pressure forces every participant to abandon standards they would prefer to keep, not because they want to, but because refusing to do so simply hands the advantage to whoever is willing to go lower.
The AI industry has been in a race to the bottom for years. The Anthropic story is the moment it became impossible to pretend otherwise.
“He (Brockman) confirmed the company started as a nonprofit to benefit all of humanity. He also confirmed his stake may be worth thirty billion dollars. He did not explain the distance between those two things.”
WHAT THIS MEANS FOR EVERY SAFETY PROMISE EVER MADE
Every major AI company has published safety commitments. OpenAI has them. Google has them. Anthropic has them. Microsoft has them. Meta has them. They come in the form of responsible use policies, safety frameworks, ethical guidelines, and public pledges.
Anthropic’s safety commitment was not vague. It was specific, no mass surveillance, no autonomous weapons. It was stated publicly. It had a named CEO attached to it. And when it cost them a customer, it lasted exactly as long as it took for the Pentagon to find someone else.
That tells you something important about the value of voluntary safety commitments in a competitive market.
It tells you they are worth exactly what they cost the company to keep them. When keeping them costs nothing, they hold. When keeping them costs a Pentagon contract, they become negotiable. When keeping them costs enough, market share, revenue, competitive position, they disappear.
This is not a prediction. It already happened. We watched it happen in real time in early 2026. Anthropic drew the line. The line moved. The business continued.
THE QUESTION NOBODY IS ASKING LOUDLY ENOUGH
If Anthropic, the company founded specifically to be the responsible alternative, the company whose CEO has spoken more seriously about AI safety than almost any other executive in the industry, if that company’s safety commitments dissolve under competitive pressure, what exactly are the rest of the safety commitments worth?
The answer is not comfortable. It is also not complicated.
Safety commitments made voluntarily, by companies operating in a competitive market, without independent enforcement, without regulatory backing, and without consequences for breaking them, are not safety commitments. They are marketing.
The real safety commitment is the one that holds when it costs something to keep it. By that standard, Anthropic held theirs longer than their competitors. They deserve credit for that. They also ultimately lost the business anyway as the market filled in around them.
The solution is not to find better companies and trust them more. The solution is to build a system where safety is not optional, where the ethical line is drawn by democratic governance with enforcement behind it, not by corporate policy with nothing behind it but good intentions.
That system does not currently exist for AI. Building it is the most important political task of this decade. And the story of the day Anthropic said no, and what happened to them for saying it, is the clearest available argument for why.
Sources: Axios (April 29 and April 30, 2026) · Federal trial testimony, Musk v. Altman, Oakland CA (April 2026) · Pentagon contract records · OpenAI founding documents (2015) · Anthropic founding statement (2021)
Cricketpocalypse is an independent channel documenting the AI industrial complex. No corporate funding. No AI company money. Just the facts and the insects.