Anthropic’s Identity Crisis: 72 Hours
TechFlow Selected TechFlow Selected
Anthropic’s Identity Crisis: 72 Hours
The endpoint of an identity crisis is often the disappearance of identity.
Navigating Web3 tides with focused insightsBy Ada, TechFlow
Tuesday, February 24. Washington, D.C., the Pentagon.
Dario Amodei, CEO of Anthropic, sat across from U.S. Secretary of Defense Pete Hegseth. According to multiple outlets including NPR and CNN, citing anonymous sources, the meeting was “polite” in tone—but its substance was anything but.
Hegseth delivered an ultimatum: By 5:01 p.m. this Friday, Anthropic must lift all restrictions on Claude’s military use—allowing the Pentagon to deploy it for “all lawful purposes,” including autonomous weapons targeting and domestic mass surveillance.
Otherwise, its $200 million contract would be canceled. The Defense Production Act would be invoked to forcibly requisition Anthropic’s technology. And Anthropic would be designated a “supply chain risk”—effectively blacklisted alongside adversarial entities such as those in Russia and China.
That same day, Anthropic released its third iteration of the “Responsible Scaling Policy” (RSP 3.0), quietly removing the company’s founding commitment—the most central one since its inception: *Do not train more powerful models unless safety measures are demonstrably in place.*
Also that same day, Elon Musk posted on X: “Anthropic is massively stealing training data—that’s a fact.” Community Notes on X added reporting that Anthropic had paid a $1.5 billion settlement over using pirated books to train Claude.
Within 72 hours, this self-proclaimed “soulful” AI company assumed three distinct roles: a martyr for safety, a thief of intellectual property, and a traitor to the Pentagon.
Which one is real?
Perhaps all of them.
The Pentagon’s “Comply or Get Out”
The first layer of the story is straightforward.
Anthropic is the first AI company granted Top Secret-level access by the U.S. Department of Defense. It secured the contract last summer, with a ceiling of $200 million. OpenAI, Google, and xAI soon followed, each receiving contracts of comparable scale.
According to Al Jazeera, Claude was deployed in a U.S. military operation this past January—reportedly involving the abduction of Venezuelan President Nicolás Maduro.
Yet Anthropic drew two red lines: no support for fully autonomous weapons targeting, and no support for mass surveillance of U.S. citizens. Anthropic maintains that AI reliability remains insufficient for weapon control—and that no legal framework yet governs AI’s use in mass surveillance.
The Pentagon isn’t buying it.
Last October, White House AI advisor David Sacks publicly accused Anthropic on X of “weaponizing fear to capture regulators.”
Competitors have already knelt. OpenAI, Google, and xAI all agreed to let the military use their AI for “all lawful scenarios.” Musk’s Grok was just approved this week for integration into classified systems.
Anthropic is the last one still standing.
As of publication, Anthropic’s latest public statement says it has no intention to yield. But the Friday 5:01 p.m. deadline looms.
An anonymous former liaison between the Justice Department and the Pentagon told CNN: “How can you simultaneously declare a company a ‘supply chain risk’ and compel it to work for your military?”
A fair question—but outside the Pentagon’s scope of concern. What matters is that if Anthropic refuses to comply, it will face coercion—or become a pariah in Washington.
“Distillation Attacks”: A Slap-in-the-Face Accusation
On February 23, Anthropic published a sharply worded blog post accusing three Chinese AI firms of launching “industrial-scale distillation attacks” against Claude.
The defendants: DeepSeek, Moonshot AI, and MiniMax.
Anthropic alleged they used 24,000 fake accounts to conduct over 16 million interactions with Claude—systematically extracting its core capabilities in agent reasoning, tool use, and programming.
Anthropic labeled this a national security threat, warning that distilled models “are unlikely to retain safety guardrails” and could be exploited by authoritarian governments for cyberattacks, disinformation, and mass surveillance.
The narrative was flawless—and perfectly timed.
It came right after the Trump administration eased chip export controls to China—and just when Anthropic needed ammunition to lobby for stricter controls.
Then Musk fired a shot: “Anthropic is massively stealing training data—and paid billions in settlements for it. That’s a fact.”
Tory Green, co-founder of AI infrastructure firm IO.Net, remarked: “You train your model on data scraped from across the internet—and then call it a ‘distillation attack’ when others learn from your public API?”
Anthropic calls distillation an “attack,” yet it’s standard industry practice. OpenAI uses it to compress GPT-4. Google uses it to optimize Gemini. Even Anthropic itself does it. The only difference? This time, Anthropic is the one being distilled.
As Erik Cambria, AI professor at Nanyang Technological University in Singapore, told CNBC: “The line between legitimate use and malicious exploitation is often blurry.”
The irony runs deeper: Anthropic just paid a $1.5 billion settlement for training Claude on pirated books. It trains on the entire web—then accuses others of learning from its public API. This isn’t double standards—it’s triple.
Anthropic set out to play the victim—and ended up exposed as the defendant.
Dismantling Safety Promises: RSP 3.0
On the very same day it clashed with the Pentagon and traded blows with Silicon Valley, Anthropic unveiled its third Responsible Scaling Policy.
Jared Kaplan, Anthropic’s chief scientist, told reporters: “We believe halting AI model training helps no one. In the context of rapid AI advancement, unilaterally committing to pauses—while competitors charge ahead—is meaningless.”
In other words: If others won’t play by the rules, we won’t pretend to either.
The core of RSP 1.0 and 2.0 was a hard-line pledge: Pause training if model capabilities outpace the reach of safety measures. That promise earned Anthropic unique credibility within the AI safety community.
RSP 3.0 deleted it.
In its place is a more “flexible” framework—one that separates Anthropic’s own safety actions from broader, industry-wide safety recommendations into two parallel tracks. It promises quarterly or biannual risk reports, subject to external expert review.
Sounds responsible?
Chris Painter, an independent reviewer from the nonprofit METR, said after reviewing an early draft: “This signals Anthropic believes it must enter ‘triage mode,’ because its methods for evaluating and mitigating risk are falling behind capability growth. It’s further proof society is unprepared for AI’s potential catastrophic risks.”
According to TIME, Anthropic spent nearly a year internally debating this rewrite—approved unanimously by CEO Amodei and the board. The official rationale: The original policy aimed to catalyze industry consensus—but the industry never caught up. The Trump administration adopted a hands-off stance toward AI development, even attempting to roll back state-level regulations. Federal AI legislation remains distant. Though building a global governance framework seemed plausible in 2023, three years later, that door is clearly closed.
An anonymous researcher who has long tracked AI governance put it bluntly: “The RSP is Anthropic’s most valuable brand asset. Removing the pause commitment is like an organic food company quietly tearing ‘organic’ off its packaging—and telling you its new testing is more transparent.”
Identity Fracture at a $380 Billion Valuation
Early February, Anthropic closed a $30 billion funding round at a $380 billion valuation—with Amazon as the anchor investor. Since inception, it has achieved $14 billion in annualized revenue. Over the past three years, that figure has grown more than tenfold each year.
Meanwhile, the Pentagon threatens to blacklist it. Musk publicly accuses it of data theft. Its core safety commitment has been scrapped. Mrinank Sharma, Anthropic’s head of AI safety, resigned—and posted on X: “The world is in danger.”
Contradictory?
Perhaps contradiction is in Anthropic’s DNA.
The company was founded by ex-OpenAI executives who feared OpenAI was moving too fast on safety—only to build a company that develops ever-more-powerful models at an even faster pace, while warning the world how dangerous those models are.
Its business model can be summed up in one sentence: *We’re more afraid of AI than anyone else—so pay us to build it.*
This narrative worked flawlessly in 2023–2024. AI safety was Washington’s hottest buzzword—and Anthropic its most sought-after lobbyist.
In 2026, the winds have shifted.
“Woke AI” has become a political liability. State-level AI regulation bills are being blocked by the White House. California’s SB 53—a bill backed by Anthropic—was signed into law, but federal action remains barren.
Anthropic’s safety card is sliding from “differentiating advantage” to “political liability.”
Anthropic is performing a delicate balancing act: It needs to be “safe enough” to preserve its brand—and “flexible enough” to avoid abandonment by markets and government. The problem? Tolerance on both sides is shrinking.
How Much Is the Safety Narrative Worth?
Lay the three events side-by-side, and the picture sharpens.
Accusing Chinese firms of distilling Claude reinforces the lobbying narrative for tighter chip export controls. Deleting the safety pause commitment keeps pace in the arms race. Refusing the Pentagon’s demand on autonomous weapons preserves a final veneer of moral legitimacy.
Each step is logically defensible—yet mutually contradictory.
You cannot claim that Chinese firms distilling your model endangers national security—and simultaneously scrap the commitment designed to prevent your own model from spiraling out of control. If models are truly that dangerous, caution—not acceleration—is the rational response.
Unless you’re Anthropic.
In the AI industry, identity isn’t defined by what you say—it’s defined by your balance sheet. Anthropic’s “safety” narrative is, at its core, a brand premium.
In the early days of the AI arms race, that premium had value. Investors paid higher valuations for “responsible AI.” Governments greenlit “trustworthy AI.” Customers paid more for “safer AI.”
But in 2026, that premium is evaporating.
What Anthropic now faces isn’t a binary choice of “to compromise or not”—but a hierarchy: *Whom do we compromise with first?* Concede to the Pentagon, and the brand suffers. Concede to competitors, and the safety promise collapses. Concede to investors, and both sides lose.
At 5:01 p.m. this Friday, Anthropic will deliver its answer.
But regardless of the answer, one thing is certain: The Anthropic that once stood on “We’re different from OpenAI” is becoming indistinguishable from everyone else.
The endpoint of an identity crisis is often the disappearance of identity.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
