You're missing an important part of the negotiation - Trump must benefit personally in some way. In this case, Greg Brockman has given by far the biggest single donation ($25m) to Trump's MAGA PAC in September last year.
You're assuming a lot about Elon's ability to assemble and execute a process competently. They will probably end up hiring people off this list and firing them later.
I think what is much more interesting is what OpenAI and Google will do. There's probably some threshold of signatories where the companies in question do not fire everyone when they decide they want the DoD's business, the question will be how many people have to sign to cross it... and will enough people sign.
I don't think Google would bat an eye at firing 500 people to secure a DoD contract, but would they fire 5,000?
Both the automated verification methods depend on Google servers and Google can almost certainly retrieve that data if they want to regardless of if the signers or verifiers delete it.
It's hard to deny an offer to become a millionaire in the next 3 years if you just hang tight at xAI, especially if you don't have any offers from competing AI labs. Also, LLMs are converging into an easy-to-replicate commodity. It doesn't matter much who wastes their money on you to build them.
I am sure some people can. I got pretty far in the interview process at xAI, but couldn't get anyone at other AI labs to even reply my emails, though...
(I dropped out at the end because there were issues with HR not wanting to hire in certain geos, and me not wanting to relocate to the USA or London)
That's exactly what is happening... Anthropic are choosing not to sell their technology to the government. I'm not sure what you're suggesting otherwise here.
This is a summary from Gemini of the news reporting:
Recent news reports from February 2026 indicate that a significant rift developed between Anthropic and the Department of War (Pentagon) following the capture of Venezuelan President Nicolás Maduro in January 2026.
According to a report by the Wall Street Journal (referenced by TRT World and others on February 14–15, 2026), the controversy originated when an Anthropic employee contacted a counterpart at Palantir Technologies to inquire about how Claude had been used during the raid.
Key Details of the Reports:
* Discovery of Use: Anthropic reportedly became aware that its AI model, Claude, was used in the classified military operation through its existing partnership with Palantir. This was allegedly the first time an Anthropic model was confirmed to be involved in a high-profile, classified kinetic operation.
* The Inquest: The Wall Street Journal and Semafor reported that an Anthropic staff member reached out to Palantir to ask for specifics on Claude's role. This inquiry reportedly "triggered the current crisis" because it signaled to the Pentagon that Anthropic was attempting to monitor or place "ad hoc" limits on how its technology was being used in active missions.
* The Confrontation: During a recent meeting between Anthropic CEO Dario Amodei and Defense Secretary Pete Hegseth, the inquiry to Palantir was a point of contention. Hegseth reportedly claimed Anthropic had raised concerns directly to Palantir about the Caracas raid. Amodei has since denied that the company raised objections to specific operations, characterizing the exchange with Palantir as a routine technical follow-up or a "self-serving characterization" by Palantir.
* Current Status: This friction has escalated into a public showdown. Today, Friday, February 27, 2026, reports indicate that the Trump administration has officially designated Anthropic a "supply chain risk" and ordered federal agencies to cease using Claude after the company refused to remove guardrails related to autonomous weaponry and mass domestic surveillance.
The primary reporting you are likely recalling comes from The Wall Street Journal (approx. February 14, 2026) and was later expanded upon by Semafor regarding the specific communications between Anthropic and Palantir employees.
Also, Trump's own words complaining about being forced to stick to Anthropic's terms of service:
> The Leftwing nut jobs at Anthropic have made a DISASTROUS MISTAKE trying to STRONG-ARM the Department of War, and force them to obey their Terms of Service instead of our Constitution.
Knowing the input text was two words separated by a space, I was able to use hashcat and the unix wordlist (/usr/share/dict/words) to find the solution almost immediately. It's a shame that Alex didn't find it this way on his first attempt as the two words are fairly common.
The 2.5kW figure is for a server running 10 HC1 chips:
> The first generation HC1 chip is implemented in the 6 nanometer N6 process from TSMC. ... Each HC1 chip has 53 billion transistors on the package, most of it very likely for ROM and SRAM memory. The HC1 card burns about 200 watts, says Bajic, and a two-socket X86 server with ten HC1 cards in it runs 2,500 watts.
That's not evidence that the model is assuming anything, and this is not a brainteaser. A brainteaser would be exactly the opposite, a question about walking or driving somewhere where the answer is that the car is already there, or maybe different car identities (e.g. "my car was already at the car wash, I was asking about driving another car to go there and wash it!").
If the LLM were really basing its answer on a model of the world where the car is already at the car wash, and you asked it about walking or driving there, it would have to answer that there is no option, you have to walk there since you don't have a car at your origin point.
reply