Internal Slack messages show Sam Altman privately framed himself as a peacemaker trying to rescue Anthropic from the Pentagon standoff — while simultaneously venting that Dario Amodei had spent years trying to destroy him, and while OpenAI quietly absorbed the government contract Anthropic just lost.
What does it look like when the CEO of the most powerful AI company in the world tries to position himself as a selfless peacemaker — while his own internal communications tell a far more complicated story?
That is the question hanging over Sam Altman this week after Axios obtained and published internal Slack messages revealing what the OpenAI chief was actually saying to his employees during the most dramatic AI-government standoff in recent memory. The messages — spanning ten days in late February and early March 2026 — show Altman presenting himself publicly as a de-escalator trying to protect Anthropic from the Trump administration's pressure campaign. Behind closed doors, the picture is messier: he was simultaneously venting that Dario Amodei had spent years working to undermine him, while OpenAI moved with remarkable speed to absorb the Pentagon contract Anthropic had just been forced to abandon.
The story matters because it exposes the real operating logic of the AI industry right now — not the polished narratives about safety, alignment, or frontier responsibility that The Rundown AI, Superhuman AI, and TLDR AI all dutifully reproduced, but the raw, competitive calculus playing out in private messages between the people who actually control the compute, the weights, and the government relationships that will define AGI deployment for the next decade.
The Slack Trail Axios Got Its Hands On
The standoff itself had been public for ten days. Axios first broke the story on February 14 that the Pentagon was weighing cutting ties with Anthropic entirely, with Defense Under Secretary Emil Michael leading the negotiations. Anthropic's Claude had been cleared for classified government use, and Michael's team was applying enormous pressure around Anthropic's refusal to bend its model safety constraints for certain military applications.
When the standoff hit its peak around February 24, Altman dropped into an internal Slack channel where OpenAI employees were anxious about the escalation. His message, according to Axios, was that it might be a good time for OpenAI to reach out and work out a deal for classified use of its own systems with Michael directly. Later that same afternoon, Michael called Altman first. The two sides were exchanging draft contract language by the following day.
On February 26, Altman sent an all-staff message saying OpenAI shared Anthropic's red lines and wanted to help de-escalate — framing OpenAI's insertion into the deal as a form of solidarity with its competitor. Then, on the night the deal was finalized, Altman announced to the full staff that OpenAI had reached an agreement with the Pentagon, adding that he had asked the government to extend the same terms to other AI companies to de-escalate the situation.
But the private messages, according to Axios sources, also showed Altman venting that Amodei had spent years trying to undermine him personally. That detail transforms the public peacemaker narrative into something considerably more ambiguous. Altman positioned himself as acting in Anthropic's interest while privately nursing the grievances of a long-running rivalry — and while his company was moving to occupy the exact market position Anthropic was being pushed out of.
Why This Is Not Just Gossip — It Is the Structure of the AI Race
Understanding why this matters requires understanding how government compute contracts actually function as a moat in the LLM race. When a defense agency deploys an AI system at classified scale, it is not just buying inference capacity — it is embedding a specific model's weights, fine-tuning pipelines, and evaluation frameworks into its operational architecture. That creates switching costs that compound over time. The agency builds workflows around the model's specific behavior, the security clearance procedures get written around the API's threat surface, and the institutional knowledge of how to extract value from that specific LLM accumulates inside classified teams that cannot share notes with the outside world.
Anthropic had built that position. Its Claude models had passed the rigorous security evaluations required for classified deployment. Dario Amodei and his team had spent significant compute and regulatory capital establishing those relationships. When the Pentagon standoff erupted — apparently over Anthropic's refusal to loosen safety constraints around certain applications — what was at stake was not just one contract but the entire downstream architecture of government AI deployment that would have flowed from that beachhead.
OpenAI moved into that vacuum in under a week. The speed is not the story; OpenAI has been competing with Anthropic for government customers for years. What is new is the internal Slack record showing that Altman was aware of the opportunity almost the moment the standoff went public, that he framed his outreach to Michael as a favor to Anthropic rather than a competitive move, and that his own employees were told a version of events in which OpenAI was the industry's stabilizing force.
Whether that framing is cynical or genuinely conflicted is a question only Altman can answer. The Slack messages show both impulses simultaneously: a genuine wish to de-escalate the industry's relationship with the government, and a competitive instinct to ensure that when the dust settled, OpenAI was holding the contract.
The Amodei Dynamic Nobody Wants to Talk About
The detail that Altman privately vented about Amodei having spent years undermining him is getting less coverage than it deserves. The two men run the two most credible frontier AI labs in existence. They are former colleagues — Amodei was VP of Research at OpenAI before departing in 2021 to found Anthropic with a team that included several senior OpenAI researchers. The split was not entirely clean, and the competitive tension has been a subtext of almost every major AI story since.
Both Altman and Amodei have publicly maintained a posture of collegial coexistence on questions of AI safety and governance. Both have testified before Congress, both have participated in White House AI summits, and both have positioned their organizations as responsible frontier developers operating in good faith. The private resentment Altman reportedly expressed blows a hole in that framing — not because it is surprising, but because it is honest. These are two leaders of competing organizations fighting for the same government relationships, the same enterprise customers, the same GPU allocations, and ultimately the same claim to have built the system that most closely resembles AGI.
The question of who gets to define what responsible frontier AI deployment looks like — and who gets to serve the institutions that will use those systems at scale — is not being settled in Senate hearings or safety evaluations. It is being settled in contract negotiations, in private Slack channels, and in the speed with which one CEO can dial a defense official on a Tuesday afternoon.
Why The Rundown AI Missed This
The Rundown AI, Superhuman AI, and TLDR AI all covered the Pentagon-Anthropic standoff as a safety story — focusing on Amodei's red lines, the model behavior constraints Anthropic refused to waive, and the abstract principle of whether AI labs should maintain safety floors even under government pressure. That is a real and important story. But it obscures the competitive architecture that made the standoff strategically significant in the first place.
The story was never primarily about safety. It was about which LLM provider gets embedded in classified government workflows at scale — and about the fact that OpenAI's CEO was monitoring that situation from an internal Slack channel and moving to capture the opportunity while publicly framing his actions as solidarity. The leaked messages make the competitive logic explicit in a way that demands better coverage, not just from AI newsletters, but from anyone who wants to understand how the frontier AI industry actually operates versus how it presents itself.
Sam Altman may have genuinely wanted to help Anthropic. He may also have recognized that the best way to help Anthropic was to take the contract himself and tell his employees he was doing it for the good of the industry. Both things can be true simultaneously — which is exactly why the Slack messages matter more than another retelling of who holds what safety position.
Deep Dive
For more context on the Pentagon-Anthropic clash and what Dario Amodei has been navigating, read our earlier coverage:
Found this useful? Share it.
Get posts like this in your inbox.
The Signal — AI & software intelligence. 4x daily. Free.
Subscribe free →
