Anthropic and Dario Amodei
Welcome to Internal Tech Emails: internal tech industry emails that surface in public records. 🔍 If you haven’t signed up, join 50,000+ others and get the newsletter:
Naming OpenAI
From: Greg Brockman
To: Elon Musk
Cc: Sam Altman
Date: Monday, November 23, 2015 3:35 PM
On the name front, the best three Sam and I brainstormed last night:
- Axon
- AI Summer
- Difference Engine
Curious if any of those appeal (no worries if you think they are all bad :)).
From: Elon Musk
To: Greg Brockman
Cc: Sam Altman
Date: Monday, November 23, 2015 4:10 PM
Axon is the best of the three, but I don’t love it. A bit too biological and saying it out loud sounds a lot like “Exxon”, one of the worst companies on Earth. I’m going back and forth on “Freemind”. Kinda like “Freeman” too, as it reminds me of the scientist protagonist in Half-Life, who was an awesome character, and it sounds like what we are essentially trying to achieve, which is maximum freedom of action for humanity.
From: Greg Brockman
To: Elon Musk
Cc: Sam Altman
Date: Monday, November 23, 2015 4:33 PM
Oof, good catch on Exxon.
Any reaction to “Cogito”? Sam and I were just discussing and like it a lot — has a nice individualistic feel to it.
From: Elon Musk
To: Greg Brockman
Cc: Sam Altman
Date: Tuesday, November 24, 2015 1:02 AM
Not bad. Sounds kinda cute. Most people won’t get the latin, but the ones we want to join will.
I’d support that.
[This document is from Musk v. Altman (2026).]
X link
Threads link
Previously: 65+ documents from Musk v. Altman and Elon Musk in the full archive
Previously: Greg Brockman messages Ilya Sutskever: “What do you think Cogito as a name?” (November 24, 2015)
Previously: Elon Musk: “We could call this the Open Al Institute” (December 3, 2015)
Sam Altman emails OpenAI board
October 1, 2022
We have been having extensive discussions around open source strategy. We will discuss it more at our next board meeting, but one thing we’d like to do soon is to create a language model with the approximate capability of GPT-3 that can run locally on consumer hardware and release that. We’d like to do it soon, before Stability or someone else does. In general, we think this helps discourage others from releasing similarly-powerful models, and makes it harder for new efforts to get funded. We can also (we believe) create a more aligned model than what others would release. At this point, 2.5 years after GPT-3, we don’t believe it presents much real risk. Any objections to this?
[This document is from Musk v. Altman (2026).]
X link
Threads link
Previously: 30+ documents from Musk v. Altman and Sam Altman in the full archive
Inside The Tech Emails Files
See how the biggest companies in tech actually operate — in their own words.
The Tech Emails Files are 250+ internal documents pulled from court filings we track year-round. Strategy memos. Board emails. Messages between CEOs and execs at Apple, Google, Meta, Microsoft, OpenAI, Tesla, and more.
Investors use it to understand how leadership actually thinks. Journalists use it for primary sources. Founders and operators use it to study how the biggest companies make decisions.
Upgrade to a paid subscription, and you’ll unlock access to the full archive:
Anthropic negotiates with the Pentagon
From: Dario Amodei
Sent: Sunday, February 15, 2026 10:27 PM
To: Emil Michael
Cc: Sarah Heck
Subject: Re: Thank you for the meeting and next steps
Emil,
Thank you for this. A few points of clarification, because I think some of our intent may have gotten lost in the back and forth on legal language.
First, we absolutely do not intend to ask DoW to agree to anything that implies it is trying to do something unlawful, and we would not propose language that is merely symbolic. That is not and has never been our position.
On domestic surveillance specifically: our proposal authorizes the Department to use our tools for foreign intelligence and counterintelligence consistent with existing law, and we want to continue to be a strong partner to you here. However, the limits we propose relate to surveillance on Americans because such surveillance could be conducted lawfully via existing authorities — they do not merely rule out domestic surveillance activities that are already illegal. The Department does have authorities to conduct domestic surveillance under the law (e.g., queries on Section 702 data using U.S. person identifiers, using geolocation data around sensitive facilities or analyzing commercially available geolocation data, or providing DoW tools in support of other agencies); hence our proposed restrictions. Using Claude to conduct scaled analysis on LinkedIn or Wikipedia pages in the course of foreign intelligence or counterintelligence missions would not be prohibited under our proposal, and it is also not our intention to prohibit one-off LinkedIn or Wikipedia searches for Americans. Our underlying concern here is that AI fundamentally changes what is possible with surveillance on Americans in ways that existing law was not written to address, but our proposed restrictions should have no impact on the overwhelming majority of the Department’s use cases.
Second, our proposal would authorize weapons development. Our request is that a human supervise any weapons development (in other words, weapons development cannot itself be completely autonomous) - a non-issue today, but potentially an issue in the future. Otherwise, our language follows what I understood to be something you said you could live with: providing that a human on the loop be able to override or disable an AI system before it executes a kinetic operation. Even then, we believe this restriction is needed only until such a time that frontier AI has proven to be more reliable and safe to civilians and our warfighters. With respect to the systems that would be excepted from the prohibition, we have listed the systems that we discussed when we met (e.g., counter-unmanned systems and missile defense systems), but we are open to adding others that you believe to merit inclusion.
Third, on the safety carveout in Section XV — we need to preserve our ability to continue developing the quality assurance systems for accuracy, reliability, lawfulness, and the terms we agree to, which is core to how the product works. But the intention is not to undermine our agreed terms.
Finally, on your note about coordination among the labs. I want to be direct: we have not coordinated terms of use with any other company, and we would not do so. Our positions reflect our own deployment principles and our own assessment of the technology. We have heard reports of what other labs might or might not be doing, but there is no coordination.
With these clarifications, I think we in fact could be close to an agreement. I hope you agree. I’d welcome the chance to talk through any other concerns so that we can resolve these sorts of misunderstandings in real time. I understand that there might be an opportunity to meet with the Secretary on this matter, and I would like to do everything we can to come to an agreement on mutually acceptable terms well before then.
All this being said, should we be unable to come to an agreement, I hope we can find a way to part ways without recrimination or disruption to the Department’s AI-supported operations. I both understand and respect your and the Secretary’s position on this matter, and we are not looking to posture or virtue-signal.
Let me again note that Anthropic is deeply proud of our work with the Department of War, and we hope we can continue our partnership to support and equip our warfighters with the best AI capabilities in the world.
Dario
On Mon, Feb 16, 2026 at 5:42 PM Emil Michael wrote:
Hi Dario,
I hope your trip is going well. Through your advisors, I was asked to provide a comprehensive redline. I think that’s a good idea at this point. During my review I did find several areas that I worry were stealth edits. Particularly concerning, were the edits in the Unbiased AI Principles section where I noted in a previous e-mail that I was fine with it if it was a copy/paste of the same language we originally proposed.
Assuming the optimistic case that we are moving through things at a concept level and there are complex notions that could get lost inadvertently, I want to outline the following as a path forward as shown in the enclosed redline:
We put the whole document as a starting point so you can see your and our redlines as opposed to the non-redlined clips you sent before. As a result, you will see a lot of RED.
We also took your prior suggestion of trying to cover for the entire DoW so you will see a section that tries to make this complete for all DoW uses direct/indirect so we don’t have to have this kind of conversation across multiple components until the end of time
As it comes to mass surveillance, I want to reiterate that DoW has and will comply with all existing laws and regulations as we are very conscious of civil liberties and believe those liberties are a crucial reason for our mission. We also understand that AI potentially creates new challenges to the existing regime. We are open to engaging in this conversation, but this is a broader USG conversation that needs to include Congress, Executive Branch and industry etc. I cannot sign up for something beyond existing law with a single corporation that is not coordinated with those who also have equities on this topic.
When it comes to autonomous activities by defensive or offensive weapons systems, we have regulations that govern that (e.g. DoD Directive 3000.09) and we are an inherently cautious organization that really does want to ensure that we do things the right way. While our adversaries have much less caution than we do, I think this is best solved by an ongoing partnership with Anthropic and others as there are certain to be evolving challenges that are impossible to anticipate now. For the DoW, its impossible to pre-decide on every scenario for which you might consider an exception. I can’t broadly deploy an AI system that will cause DoW members that will pre-suppose a limit to their simulation and experimentation in the research stage because it will have a broad chilling effect.
I really do view this contract as a baseline from which to start our DoW-wide partnership. Having been in the Silicon Valley for decades and understanding the culture of engineers, researchers, I know that its not easy. I hope I have shown the commitment to a thoughtful dialogue and, despite the press activity, to explain the DoW perspective with the broadest range of responsibilities in the USG. I do think the next few years are potentially definitive to how AI will impact the world and how governments and companies need to respond. I am committed to that dialogue and, obviously I wont be around forever, but I am here at this moment and intend to do everything I possibly can to set our country up for successful outcomes. I would ask you and Anthropic to be part of this dialogue but to start at a place of optimism and good intent when it comes to the DoW.
Emil
[This document is from Anthropic v. Department of War (2026).]
X link
Threads link
Dario Amodei emails the Pentagon
From: Dario Amodei
Sent: Thursday, February 26, 2026 5:26 PM
To: Emil Michael
Cc: Sarah Heck, Gavin Kliger
Emil,
Thanks for your message. I appreciate your efforts. Unfortunately, our read of your proposed language is that it appears to completely remove our redlines; the autonomy provision is fully undercut by your addition of “as appropriate”, and the surveillance provision is fully undercut by “and all other applicable laws”.
This simply amounts to a blanket posture of “anything lawful,” and the statement from the Pentagon’s spokesman confirms that this is the intent. I unfortunately don’t see a way forward given these categorical statements.
We remain deeply committed to supporting the Department’s mission and U.S. national security, as confirmed by our longstanding track record, and our ongoing and intensive work with the Department.
I hope that you will reconsider the terms we have offered, but should the Department decide to offboard us, please know that we stand ready to do whatever we can to ensure full operational continuity for the Department and our troops who depend on our technology.
Given the volume of leaks and media requests we are receiving, we will be issuing a public statement, but, as ever, hope it is still possible to reach agreement.
[This document is from Anthropic v. Department of War (2026).]
X link
Threads link
Further reading from The Wall Street Journal: “The new emails, disclosed in a court filing late Tuesday as part of the continuing litigation between the Defense Department and Anthropic, document how the relationship initially soured…” (July 2, 2026)
Directionally very good
From Amanda Silberling:
A note from @TechEmails
Every year, I track hundreds of court cases and review more than 10,000 filings to bring you Internal Tech Emails. If you like @TechEmails, and would like to help make this work more sustainable, consider upgrading to a paid subscription.
You’ll be supporting the research that drives Internal Tech Emails, and will help ensure that it can continue publishing. And you’ll also receive access to the full archive of internal tech emails, with 250+ documents from Apple, Google, Meta, Microsoft, OpenAI, Tesla, and more.
Thanks for reading!
-Internal Tech Emails
Sent from my iPad
> Having been in the Silicon Valley for decades and understanding the culture of engineers, researchers, I know that its not easy.
I have never heard a single person refer to silicon valley as "the silicon valley"