Two days ago we covered a story about the White House drafting guidance to walk back its supply-chain risk designation against Anthropic. The headline was "ANTHROPIC: White House Folded." It looked like the standoff was ending. Today the Pentagon clarified its position with no room for misinterpretation. The Defense Department announced classified network deals with seven major AI companies. SpaceX. OpenAI. Google. NVIDIA. Reflection. Microsoft. AWS. Anthropic was deliberately excluded. Pentagon CTO Emil Michael told CNBC directly that Anthropic remains a supply chain risk. The ban is not ending. The administration may be softening but the Pentagon is holding the line. The most powerful cybersecurity AI ever built, the one Anthropic developed and gated to 40 elite defenders, still cannot be deployed inside US Department of Defense networks. And the seven companies that the Pentagon did pick include some of Anthropic's biggest direct rivals.

Here is what the Pentagon agreed to today.

Seven AI companies signed deals. SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and AWS will all deploy AI capabilities on the Defense Department's classified networks. This expands the Pentagon's AI footprint dramatically. The agreements integrate the seven companies into the Pentagon's Impact Levels 6 and 7 environments, which are the most sensitive classified network tiers in US military infrastructure.

The Pentagon's main AI platform is already real. Gen AI has been used by over 1.3 million Defense Department personnel in just five months of operation. That is a deployment scale most enterprise AI rollouts could only dream of. The Pentagon is not experimenting with AI. The Pentagon is fully deploying it across classified workflows at massive scale.

Anthropic is the one major lab not on the list. Reuters explicitly called out the exclusion. Pentagon CTO Emil Michael told CNBC today that Anthropic is still a supply chain risk. He went on to say something interesting about Mythos specifically. He called it "a separate national security moment" because of its ability to supercharge hackers. The Pentagon is not pretending Mythos does not exist. They are saying Mythos is so powerful it requires a separate framework before it can be deployed inside DoD networks. Translation: even when the supply chain designation eventually softens, Mythos itself faces a higher bar.

The president weighed in last week. Trump said publicly that Anthropic was "shaping up" in the eyes of his administration, opening the door for the company to reverse its blacklisting at the Pentagon. That comment plus Wednesday's Axios reporting on the White House drafting workaround guidance created the impression that the standoff was ending. Today's Pentagon announcement clarifies the actual situation. The White House may be softening but the Pentagon is not. They are two different bodies with two different timelines.

Because Anthropic is still refusing to drop its usage restrictions on surveillance and autonomous weapons.

The original supply-chain designation came after Anthropic refused to allow its models to be used for certain Pentagon AI applications. Anthropic did not budge. The Pentagon imposed the ban. Six weeks later, here we are. Anthropic still has not changed its usage policy. The Pentagon still has not changed its position. Both sides are holding firm.

What changed today is the comparison set. The Pentagon was previously banning Anthropic in isolation. Today they made it explicit. Six other major AI labs were happy to sign classified deals on Pentagon terms. SpaceX. OpenAI. Google. NVIDIA. Reflection. Microsoft. AWS. The list is a public statement. The Pentagon is saying we have plenty of options. We do not need Anthropic. If you want to play, you play by our rules. Anthropic chose not to. The seven companies on this list chose otherwise.

For Anthropic this is a financial loss but not a fatal one. Federal contracts at Pentagon scale are worth hundreds of millions of dollars per company. Losing access to that pool hurts. But Anthropic just secured $65 billion in investment commitments from Amazon and Google in five business days. They have $30 billion in annualized revenue. The federal contract pool is meaningful but not existential.

For the Pentagon this is a strategic statement. They are building the most ambitious classified AI infrastructure in the world. They are doing it without Anthropic. And they are signaling to every other AI lab watching that the Pentagon will not negotiate on usage restrictions for sensitive applications.

David Sacks said something important today that did not make most headlines. The White House AI czar told reporters that Mythos is not a doomsday device. He said it is simply the first AI model capable of automating cyber tasks. He added that GPT-5.5-Cyber has already reached the same level. And he said all frontier models will reach this level within six months.

If that is true, the Pentagon's position becomes harder to defend over time. The Pentagon is keeping Anthropic out partly because Mythos is too powerful and too dangerous to deploy under current frameworks. But if every other frontier model reaches Mythos-level cyber capability within six months, the supply chain designation only delays the Pentagon's access to capabilities they will eventually have through other vendors anyway.

That is probably why the White House is drafting workaround guidance. Senior administration officials see what is coming and want to keep Anthropic's most capable tools available to defenders. The Pentagon is operating on a slower timeline because their procurement and risk frameworks are different.

The likely path forward is messy. Anthropic continues to hold its line on surveillance and autonomous weapons. The White House finds a workaround that gives federal civilian agencies access to Mythos through a separate framework. The Pentagon eventually creates a "national security moment" framework specifically for Mythos-class models. By the time that happens GPT-5.5-Cyber and other frontier models will probably qualify too. The exclusion stops being about Anthropic specifically and becomes about which class of cyber AI any agency can deploy at all.

The Pentagon announced classified AI deals today with SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and AWS. Anthropic was not included. Pentagon CTO Emil Michael confirmed that Anthropic remains a supply chain risk. The White House is drafting workaround guidance for civilian agencies but the Pentagon is holding firm. Anthropic refused to drop its usage restrictions on surveillance and autonomous weapons. The Pentagon refused to give them an exception. Six weeks into the standoff both sides are still standing. Anthropic just raised $65 billion from Amazon and Google. The Pentagon just signed seven competing AI labs. Neither side needs the other to survive. The Banned In America storyline is not over. It is entering a longer phase where two sides who do not need each other are testing how long they can hold the line. Tonight the Pentagon held theirs.

Stay building. 🤖