Just now, Ultraman showed off: We have signed a contract with the Pentagon, and our revenue exceeds 20 billion. Overnight, OpenAI was criticized by thousands of people, and a wave of ChatGPT uninstallation started across the entire network! Yesterday, Ultraman announced a cooperation with the US Department of Defense (now renamed Department of War, DoW).
Anthropic rejects the use of AI for large-scale digital surveillance and automated weapons, and OpenAI adds insult to injury and quickly takes over similar contracts.
In contrast, OpenAI is quite shameless. The United States spontaneously launched a movement to uninstall/cancel ChatGPT, which was huge and popular!
OpenAI first announced the details of the cooperation in an attempt to restore its reputation, but public opinion surged. Altman and other OpenAI executives answered relevant questions online in real time, and more than 6 million netizens watched:
Not all OpenAI employees agree with this cooperation, but Ultraman and others are too embarrassed and too anxious. Ultraman frankly said: "I don't know how to end it."
They thought they were smart, but they opened Pandora's box with their own hands, turning AI from a human assistant into a weapon of war.
But of course OpenAI is not the only one doing this.
Amid internal and external troubles, OpenAI employees reveal their past
After Anthropic was blocked by Trump, it immediately won sympathy and support from the public, including AI researchers from AI giants such as OpenAI and Google.
In order to dispel internal and external doubts, relevant OpenAI executives argued:
(The transaction amount between OpenAI and the U.S. Department of Defense) is only a few million dollars, which is insignificant compared to our annual revenue of more than 20 billion U.S. dollars. It is definitely not worth the public relations controversy caused by this.
We do it because it's the right thing for the country, at a huge cost to ourselves, and never for the revenue.
This time, OpenAI seems to have picked up the sesame seeds and lost the watermelon, and the gain is completely outweighed by the loss!
In fact, many people believe that this is OpenAI’s “signature letter” to the White House:
The strategic value of being selected to the U.S. government far exceeds any dollar figure on the contract. A few million dollars can help you win the title of "American AI Company" on the eve of a trillion-dollar IPO.
Through this contract with the Department of Defense, OpenAI’s fundamental purpose is to be deeply tied to the U.S. government to gain potential advantages and benefits!
This is not the first time OpenAI has compromised for the sake of profit.
Some OpenAI employees have already seen through the true nature - they voted with their feet and left OpenAI.
It’s not just sentiment, it’s a values vote.
01
19-year-old genius employee quits job angrily
At the beginning, he joined OpenAI full of ideals, not to make weapons, but because he believed that OpenAI was committed to developing AI that could benefit all mankind——
Instead of taking advantage of Anthropic and becoming the "arms contractor" who succeeded them after they were blacklisted for adhering to their principles.
And the situation is worse than he thought:
OpenAI allows the Department of Defense to use their models for "all lawful purposes."
"Legal" is the bottom line, not conscience. conscienceA higher threshold is needed.
So, this time he completely broke his guard and his dream was shattered:
Altman says the security guarantee is real.
Maybe. But I’ve seen this industry promise time and time again.
He has seen "we will never" become "under certain circumstances" and then "the government has agreed in principle".
"We will never develop independent weapons." "We will never participate in domestic surveillance."——
These were also the red lines that Anthropic adhered to, and it was precisely because they adhered to these bottom lines that they were labeled a "national security risk."
OpenAI's AI model is deployed to confidential military networks, and it is difficult to know what those confidential networks will ultimately be used for: "I don't know, you don't know. Ultraman doesn't know either."
This uncertainty is the key, and the 19-year-old cannot make peace with it and hopes to spend his future time in a place where he does not need to explore these terrible answers.
So, he accepted the position at Anthropic.
What this means, he knows very well:
Anthropic is now blacklisted, embroiled in lawsuits, and fighting a government that calls them "left-wing lunatics" because they believe humans should be in control of weapons.
That's exactly why he joined Anthropic.
Some things are more important than contracts.
02
Former OpenAI employee reveals secrets
Sarah Shoker, a senior research scholar in human AI at the Berkeley Risk and Security Laboratory at the University of California, Berkeley, and former head of the OpenAI geopolitics team, provided high-value observations from first-hand information within the OpenAI policy team.
After working on OpenAI’s geopolitics team for about three years, she moved to two other teams before finally deciding to leave in June 2025.
As a researcher, she was often asked to provide "feedback" on military use policy.
Over the past five years, she has watched OpenAI update its usage policy several times.
In 2021, when she first joined the company, OpenAI completely banned the military use of AI.
This ban changes in 2024. But the wording of the new rules is, "to put it mildly, vague and evasive."
What she wants to ask this time is very simple: OpenAI please don’t be vague and perfunctory everyone and employees.
Let’s jump right in to a recent report from Bloomberg, which revealed that OpenAI’s models are being used in drone swarm trials.
On February 12, an OpenAI spokesperson told Bloomberg that in trials used by the Defense Innovation Unit and the Defense Autonomous Operations Group to build drone swarms, they used open source models and needed to comply with the company's usage policy.
It turns out that this usage policy can be interpreted in various ways, depending on whether you believe using AI voice-to-digital tools in the kill chain is equivalent to helping build weapons, or whether you believe an AI model can be treated separately from its larger weapons system.
What conclusions should we, as a public, draw from such assurances?
The public needs to keep their eyes open and carefully scrutinize the public information released by cutting-edge AI companies.
A quick tip is to pay special attention when a company argues that its technology is not used for "offensive" or "kinetic" purposes.
In 2018, Google played a similar PR trick when surrounding the smart drone Project Maven contract.
Project Maven can identify adversaries by collecting data from drones and satellite sensors
At the time, Google claimed that its object detection software was "non-offensive."
One interpretation is that this was an attempt to quell criticism from employees and the public.
Another interpretation is that this is a serious misunderstanding of the distributed power characteristics of humans and machines working together in war.
They are called autonomous weapons systems. In the AI weapon system, the part with the least AI component is the trigger itself.
Now, OpenAI has accepted the wording "for all lawful purposes" requested by the War Department, and the US military has used Claude air strikes!
The biggest losers in all this are ordinary people and civilians in conflict zones.