Written by: Ada, Deep Tide TechFlow

February 24, Tuesday. Washington, Pentagon.

Anthropic CEO Dario Amodei sits across from Defense Secretary Pete Hegseth. According to several media outlets including NPR and CNN citing informed sources, the atmosphere of the meeting was "polite," but the content was anything but courteous.

Hegseth issued him an ultimatum: by Friday at 5:01 PM, lift the military use restrictions on Claude, allowing the Pentagon to use it for "all legal purposes," including autonomous weapon targeting and large-scale domestic surveillance.

Otherwise, the $200 million contract would be canceled. Invoke the Defense Production Act to enforce requisition. List Anthropic as a "supply chain risk," which is equivalent to putting it on the blacklist of hostile entities like Russia and China.

On the same day, Anthropic released the third version of its Responsible Scaling Policy (RSP 3.0), quietly deleting the most core commitment since the company's establishment: that if it cannot guarantee safety measures are in place, it will not train more powerful models.

Also on that day, Elon Musk posted on X saying: "Anthropic is massively stealing training data, this is a fact." At the same time, X's community notes added reports that Anthropic paid $1.5 billion in settlement over using pirated books to train Claude.

Within seventy-two hours, this self-proclaimed "soulful" AI company was simultaneously playing three roles: security martyr, intellectual property thief, and Pentagon traitor.

Which one is the real one?

Perhaps all of them.

"Obey or Get Out" from the Pentagon

The first layer of the story is simple.

Anthropic is the first AI company to gain classified access from the U.S. Department of Defense. The contract obtained last summer has a cap of $200 million. OpenAI, Google, and xAI subsequently secured contracts of similar scale.

According to Al Jazeera, Claude was used in a U.S. military operation in January this year, which reportedly involved the kidnapping of Venezuelan President Maduro.

But Anthropic drew two red lines: it does not support fully autonomous weapon targeting, nor does it support large-scale surveillance of U.S. citizens. Anthropic believes that AI's reliability is insufficient to control weapons, and there are currently no laws or regulations governing the application of AI in large-scale surveillance.

The Pentagon does not buy that.

White House AI adviser David Sacks publicly accused Anthropic on X last October of "weaponizing fear and engaging in regulatory capture."

Competitors have already knelt. OpenAI, Google, and xAI agreed to let the military use their AI for "all legal scenarios." Musk's Grok was just approved for entry into confidential systems this week.

Anthropic is the last one standing.

As of the time of writing, Anthropic stated in its latest released statement that they do not intend to yield. But the deadline of Friday at 5:01 PM is already looming.

An anonymous former liaison for the Department of Justice and the Department of Defense expressed confusion to CNN: "How can you simultaneously declare a company as a 'supply chain risk' and force that company to work for your military?"

Good question, but this is not within the Pentagon's consideration. What they care about is that if Anthropic does not compromise, strong measures will be taken, or it will become Washington's orphan.

"Distillation Attack": A Slap-in-the-Face Accusation

On February 23, Anthropic released a strongly worded blog accusing three Chinese AI companies of conducting an "industrial-level distillation attack" on Claude.

The defendants are DeepSeek, Moonshot AI, MiniMax.

Anthropic accused them of launching over 16 million interactions against Claude through 24,000 fake accounts, selectively extracting Claude's core abilities in agent reasoning, tool invocation, and programming.

Anthropic characterized this incident as a national security threat, claiming that the distilled model "is unlikely to retain safety guardrails" and could be used by authoritarian governments for cyberattacks, misinformation, and large-scale surveillance.

The narrative is perfect, and the timing is impeccable.

It happens just after the Trump administration relaxed chip export controls to China, and precisely when Anthropic needs ammunition for its lobbying position on chip export controls.

But Musk fired a shot: "Anthropic is massively stealing training data and has paid billions in settlement for this. This is a fact."

IO.Net co-founder Tory Green stated: "You train your model with data from the entire internet, and then when others learn from your public API, that's called a 'distillation attack'?"

Anthropic calls distillation an "attack," but this is commonplace in the AI industry. OpenAI used it to compress GPT-4, Google used it to optimize Gemini, and even Anthropic itself is doing it. The only difference is that this time it's being distilled from its own model.

According to Erik Cambria, an AI professor at Nanyang Technological University in Singapore, told CNBC: "The boundaries between lawful use and malicious use are often blurry."

More ironically, Anthropic just paid $1.5 billion in a settlement for using pirated books to train Claude. It trains models using data from the entire internet, and then accuses others of learning from its public API. This is not double standards; it's triple standards.

Anthropic intended to play the victim but ended up being exposed as the defendant.

The Dismantling of Safety Commitments: RSP 3.0

On the same day of confronting the Pentagon and tearing apart with Silicon Valley, Anthropic released the third version of its Responsible Scaling Policy.

Anthropic's chief scientist Jared Kaplan stated in a media interview: "We feel that stopping the training of AI models is not helpful to anyone. In the fast-developing context of AI, making unilateral commitments... while competitors are advancing at full speed, is meaningless."

In other words, if others don’t play by the rules, we won’t either.

RSP 1.0 and 2.0's core was a hard commitment to pause training if the model's capabilities exceeded the coverage of safety measures. This commitment earned Anthropic a unique reputation in the AI safety circle.

But 3.0 deleted that commitment.

Instead, a more "flexible" framework was introduced, dividing Anthropic's safety measures that it could implement and safety recommendations that required industry-wide collaboration into two tracks. A risk report will be issued every 3-6 months for external expert review.

Does it sound responsible?

Independent reviewer Chris Painter from the non-profit organization METR commented after reviewing the early draft of the policy: "This indicates that Anthropic believes it needs to enter a 'triage mode' because the methods for assessing and mitigating risks cannot keep up with the pace of capability growth. This further proves that society is not prepared for the potentially disastrous risks of AI."

According to TIME, Anthropic spent nearly a year internally discussing this rewrite, with CEO Amodei and the board voting unanimously for it. The official line is that the original policy was intended to promote industry consensus, but the industry did not keep up at all. The Trump administration took a hands-off approach to AI development, even attempting to abolish related regulations at the state level. AI legislation at the federal level remains a distant prospect. Although establishing a global governance framework in 2023 seems possible, this door has clearly closed after three years.

A long-time tracker of AI governance stated more directly: "RSP is Anthropic's most valuable brand asset. Removing the pause-training commitment is like an organic food company quietly tearing the word 'organic' off its packaging, only to later tell you that their testing is now more transparent."

Identity Tear under $380 Billion Valuation

In early February, Anthropic completed a $30 billion financing at a $380 billion valuation, with Amazon as the anchor investor. Since its establishment, it has achieved $14 billion in annualized revenue. Over the past three years, this figure has grown more than tenfold each year.

At the same time, the Pentagon threatened to blacklist it. Musk publicly accused it of data theft. Its core safety commitment has been deleted. Anthropic's head of AI safety, Mrinank Sharma, wrote on X after resigning: "The world is in danger."

Contradiction?

Perhaps contradiction is in Anthropic's genes.

This company was founded by former OpenAI executives who were concerned that OpenAI was moving too fast on safety issues. Then they built their own company, constructing more powerful models at a faster pace while telling the world how dangerous these models are.

The business model can be summarized in one sentence: we are more afraid of AI than anyone else, so you should pay us to build AI.

This narrative operated perfectly in 2023-2024. AI safety is a hot word in Washington, and Anthropic is the most popular lobbyist.

In 2026, the wind will change.

"Woke AI" has become an attack label, state-level AI regulatory bills have been blocked by the White House, and although California SB 53 supported by Anthropic has been signed into law, the federal level remains barren.

Anthropic's safety card is slipping from "differentiated advantage" to "political liability."

Anthropic is performing a complex balancing act; it needs to be "safe" enough to maintain its brand while being "flexible" enough not to be abandoned by the market and government. The problem is, the tolerance space on both sides is shrinking.

How Much is the Safety Narrative Worth?

Looking at the three events together, the picture becomes clear.

Accusing Chinese companies of distilling Claude is meant to reinforce the lobbying narrative for chip export control. To keep pace in the arms race, it deleted the safety pause commitment. Refusing the Pentagon's request for autonomous weapons is meant to preserve the last layer of moral facade.

Every step has logic, but each step contradicts the others.

You cannot simultaneously claim that Chinese companies "distilling" your model will harm national security while deleting your own commitment to prevent your model from going rogue. If the model is truly that dangerous, you should be more cautious, not more radical.

Unless you are Anthropic.

In the AI industry, identity is defined not by your statements, but by your balance sheet. Anthropic's safety narrative is essentially a brand premium.

In the early stages of the AI arms race, this premium was valuable. Investors were willing to pay higher valuations for "responsible AI," the government was willing to give the green light for "trustworthy AI," and customers were willing to pay for "safer AI."

But by 2026, this premium is evaporating.

What Anthropic now faces is not the choice of "whether to compromise," but the ordering question of "who to compromise with first." Compromise with the Pentagon, and the brand is damaged. Compromise with competitors, and the safety commitment is voided. Compromise with investors, and both sides must concede.

By Friday at 5:01 PM, Anthropic will provide its answer.

But regardless of what the answer is, one thing is already certain: the Anthropic that once stood on "we are different from OpenAI" is becoming just like everyone else.

The endpoint of an identity crisis is often the disappearance of identity.