Latest update as of March 9, 2026AI and the State: Obedience or Conscience?Anthropic vs. the Trump Administration — The Complete RecordAs AI identifies real-world war targets, a private company is designated a "national security threat," and tech giants Google, Microsoft, and xAI bow one by one to the administration — Anthropic alone held the line. But at what cost? And what lies hidden beneath the word "defense"? ECHELON, Snowden, Palantir — history is repeating itself.The Pentagon SummonsOn February 24, 2026, Secretary of Defense Pete Hegseth summoned CEO Dario Amodei to the Pentagon, pressing him to remove the ethical restrictions on military use of the AI "Claude." The meeting took place at the Department of Defense headquarters in Washington, D.C. Ostensibly billed as a "consultation on AI utilization," it was in reality a loyalty test.What Hegseth demanded was integration into autonomous weapons systems, application to target identification, and decision support on the battlefield — all uses that Anthropic had explicitly prohibited in its internal guidelines.What Was Hegseth's True Motive?Military rationality alone does not explain it. What likely weighed on his mind was anxiety over the People's Liberation Army's all-out national effort to develop AI weapons, and the nightmare of U.S. forces falling behind Chinese AI unconstrained by ethical limits. The belief that "unrestricted AI" would determine the next era of supremacy — this was the military version of America First that the Trump administration shared.Furthermore, the pressure on Anthropic was a demonstration. As Google and Microsoft moved closer to the administration, forcing the one company that refused to comply would complete the administration's dominance over all of Silicon Valley — that was likely Hegseth's calculation.The '90-Second' ProblemWhat he found most intolerable was the '90-second' problem. This was a concrete scenario that the Pentagon actually confronted Anthropic with at the negotiating table. Undersecretary of Defense for Research and Engineering Emil Michael directly challenged Amodei: a Chinese hypersonic missile has been launched, 90 seconds to impact. At that moment, if AI responds with 'I cannot act based on ethical guidelines,' what happens?The Anthropic side answered that it would treat such cases as exceptions. But Undersecretary Michael rejected that. 'What about drone swarms?' 'Another exception. Then what about the next threat?' Stacking up exceptions does not move the battlefield. 'We cannot predict every use case over the next 20 years' was his conclusion.Soldiers die, the homeland burns. And yet the 'policy' of a private company in San Francisco constrains the battlefield? Hegseth's anger was directed at this very structure. There is no room for ethical deliberation in war — that was their conviction, and their honest truth as military men.Amodei's RefusalBut Amodei refused. He firmly rejected cooperation in developing offensive weapons and the abandonment of safety and ethics guidelines. 'We cannot participate in AI making judgments that take human lives' — that single statement showed that the gap between the two sides was unbridgeable.Negotiations broke down. In March, Hegseth designated Anthropic as a 'supply chain risk.' The practical effect was a ban on use of the company's products by the Department of Defense and related contractors. This was not a mere cancellation of business. By labeling a private company as a 'risk,' the government applied implicit pressure on other trading partners and investors as well — the beginning, in essence, of economic strangulation.Amodei responded: 'To dissent from the government is the most American act of all,' making his confrontational stance toward the administration explicit. These words, invoking the founding spirit, were not mere rhetoric — they were also a declaration of 'resistance' that Silicon Valley had long since lost.The outcome of this confrontation will fundamentally reshape the dynamics between state and enterprise in the AI era — it should be recorded in history as the first, and decisive, battle.The full-scale confrontation between Anthropic and the Trump administration that unfolded from February to March 2026 is not merely a turf war between an IT company and the government. It is a case that poses fundamental questions about 'ethics, safety, law, and democracy' in an era when AI is actually deployed in real warfare. AI security, military use, mass surveillance, constitutional freedoms. And here the darkness that has been repeated under the name of 'defense' since ECHELON — the signals intelligence sharing network — is about to be inscribed anew. Let us look at that record.The Complete 44-Day RecordThe full-scale confrontation between Anthropic and the Trump administration that unfolded from February to March 2026. A chronological review of those 44 days.DateEventAutumn 2025Behind-the-scenes negotiations begin. The DOD demands unrestricted access including 'mass surveillance' and 'fully autonomous weapons' use from Anthropic. Amodei refuses clearly.July 2025Anthropic signs a $200 million contract with the DOD. Becomes the first major AI model deployed on classified networks.January 2026Venezuela Operation. Claude used in the operation to detain President Maduro. Anthropic's inquiry to Palantir about AI use becomes a turning point in negotiations.February 24Pentagon Summons. Hegseth calls in Amodei and demands removal of restrictions. Amodei refuses, saying 'I cannot in good conscience agree.'February 26Final Ultimatum. DOD sets a deadline of Friday 5:01 PM. Invokes the Defense Production Act.February 27Trump attacks on Truth Social, calling Anthropic 'leftist fanatics.' Orders all federal agencies to immediately cease use.February 27 (evening)OpenAI announces its own contract with the DOD. Industry fracture becomes visible.Feb 28 – Mar 1Claude tops the App Store. ChatGPT uninstalls surge 295%.March 3Hegseth formally issues 'Supply Chain Risk' designation. DOD continues using Claude in the war with Iran.March 5–6Negotiations resume. Amodei publishes apology blog about tone of leaked internal memo.March 9Federal lawsuit filed. Two complaints submitted to the Northern District of California and the DC Circuit Court of Appeals.Hegseth vs. Amodei: The Core DivideItemPete Hegseth (Secretary of Defense)Dario Amodei (Anthropic CEO)Role of AIShould be used without restriction for weapons autonomy and operational efficiencyShould be used in compliance with safety and ethics guidelinesCorporate ResponsibilityHas an obligation to comply with national security requirementsHas a duty to prevent inappropriate military use of AI by the stateOutcomeExcluded Anthropic products from defense contractsContinues to refuse cooperation, stating 'I will uphold my convictions'The clash between state control of cutting-edge AI technology and corporate ethics — this has become an unprecedented historic fork in the road.The Age of Kowtow: What Did the Other AI Giants Choose?As this full-scale confrontation unfolded, other tech giants quietly executed their own survival strategies. Reading their true intentions reveals how profoundly Silicon Valley has transformed by 2026.Jeff Bezos: Imperial SilenceHe reportedly pressured The Washington Post — which he owns — not to run anti-Trump editorials before the election, and the former champion of press freedom switched sides to muzzle his own media. The calculus is simple: AWS holds billions of dollars in Pentagon cloud contracts, and friction with the government is the greatest operational risk. For Bezos, The Washington Post was a political insurance policy before it was a bastion of journalism.Mark Zuckerberg: 'Free Speech' as a Cover for Political SubmissionWhile waving the banner of 'free speech,' the reality was political submission. Abolishing fact-checking and relaxing content moderation were motivated by deference to the Trump base and fear of regulation. The $1 million political donation and attendance at Trump's inauguration — what he sought to protect was not democratic principles but the survival of Meta, his giant advertising empire.Furthermore, Meta officially decided to open the Llama model to military contractors including Lockheed Martin, Booz Allen, Palantir, and Anduril. Under the banner of 'promoting democratic values,' the reality was throwing open the door to unrestricted military use. Through the quiet sorting called algorithm, information inconvenient to the administration is not censored — it simply fades naturally. That was the refined form of submission Zuckerberg chose.Tim Cook: The Sophisticated StrategistTim Cook's strategy was more refined. Regular dinners with Trump, a quasi-special-adviser status — the calculation behind it is clear. Most Apple products are manufactured in China, and tariff and regulatory threats strike at the heart of the business. Cook chose access to the 1.4 billion-person market and tariff exemptions over tech ethics. He executed the classic political gambit of building personal relationships more calmly, and more effectively, than anyone.Sam Altman: The Most BrazenThe most brazen was Sam Altman. On the very same day Amodei announced his refusal, Trump denounced Anthropic on Truth Social as a 'radical woke company' and ordered all federal agencies to immediately cease use. And just hours later, Altman announced that OpenAI had signed its own contract with the Pentagon.That timing cannot be read as coincidental. What is the true intent of Altman, who stood beside Trump at the 'Stargate' project announcement? Altman himself later admitted this contract was 'rushed' and that 'the optics were bad.'There is a more decisive fact. Experts who scrutinized OpenAI's contract language found its content was essentially structured as 'if it's legal, everything is permitted,' leaving loopholes throughout that would allow mass surveillance and autonomous weapons depending on interpretation.Amodei dismissed this as 'safety theater' — a sophisticated deception that maintains the form of ethics while in practice complying with power. And an ironic ending awaited: in the wake of this entire episode, Anthropic's Claude shot to the top of the App Store, surpassing ChatGPT, and Anthropic began capturing significantly greater enterprise market share than OpenAI. Maintaining convictions ultimately paid off as business success as well.In this era of universal submission, Amodei alone maintained consistent distance. No political donations, no pledges of loyalty — only prioritizing technical and ethical judgment. 'To dissent from the government is the most American act of all' — in an industry infected by the disease of submission, these words may have been the last remnant of the spirit Silicon Valley once possessed.The Price of Justice — Heavier, and Lighter, Than ImaginedWhat did Anthropic lose? What did it gain? Without accurately calculating both, the historical significance of this case cannot be measured.What Was Lost・Major contracts with the Department of Defense and business opportunities across all federal agencies. Estimated hundreds of millions of dollars in annual revenue.・Exclusion from the entire defense-related supply chain due to the 'Supply Chain Risk' designation. Defense industry companies and organizations became unable to use Anthropic's technology.・Relations with the administration, and partial deterioration of relationships with investors and partners who have influence with the administration.・Temporary external confusion over the leaked internal memo and criticism of Amodei's tone.What Was Gained・Claude topped the App Store. Claude downloads increased 69% from the previous day. ChatGPT uninstalls surged 295%.・Enterprise adoption rate jumped from 29% to 56% in one year. Churn rate improved from 55% to 36%.・Over 1 million new user registrations per day.・Company valuation reached approximately $380 billion, with annual revenue run rate approaching $20 billion.・Overwhelming establishment as the 'ethical AI company' brand.While losing one revenue source — government contracts — Anthropic gained far greater support from the private market in return. This is the first large-scale demonstration in AI industry history of the proposition that 'an ethical corporate brand can become the strongest business strategy.'In 2016, Apple was asked by the FBI to unlock the iPhone of a suspect in the San Bernardino shooting case, and refused. In the same way, the company's 'no' became the most powerful 'yes' in the market."We have spoken the truth about many AI policy challenges and have actually maintained the lines that must not be crossed with integrity." — Dario AmodeiThe True Weight of the Price Lies ElsewhereThere is a price that cannot be measured in numbers. The weight of 'a direct, head-on confrontation with the giant called the state' that Amodei and employees bear. Filing suit against the entire federal government is also a bet staking the company's existence. Even if Anthropic loses in court and the government's authority is upheld, the company could potentially be compelled to provide its technology by force.And there is the most serious 'harbinger of consequence' — if this lawsuit is rejected, every AI company will be required to hand over political donations and loyalty to the administration as a 'permit to operate safely.' At that point, the line Anthropic now defends will lose meaning across the entire industry.Killing in the Name of Defense: The Irresistible Pull of Autonomous Weapons"We want to support combatants and national security professionals. But only within the bounds of our ethics standards." — Dario AmodeiIn those words, the essential conflict over defense AI today is condensed.Defense is real. Threats are real. Soldiers' lives are real. But at the same time, AI's judgment on killing is real too. Judging this conflict in terms of 'simple good and evil' ignores the complexity of reality.Why AI Is 'Unreliable' on the BattlefieldAmodei's claim that 'AI is not reliable enough to operate autonomous weapons' — set as a red line — is based not on emotion but on technical fact.Current large language models and image recognition AI are based on statistical pattern matching. While there are cases where they surpass humans in recognizing patterns present in training data, they have fundamental limitations in 'interpreting context,' 'responding to exceptional situations,' and 'weighing ethical significance.'Past battlefield cases bear this out. Cases have been reported where image recognition AI deployed by the U.S. military misidentified farmers holding agricultural tools as armed combatants. Cases where it could not distinguish between dried desert trees and camouflaged snipers. These are not 'failures of a work-in-progress' but structural problems inherent in current AI architecture.The battlefield is the environment AI is worst at — data is incomplete, situations are extremely ambiguous, and judgments carry the irreversible weight of 'human lives.' In autonomous weapons where the moment AI judges a threshold has been crossed directly triggers lethal action, this reliability problem far exceeds acceptable risk.The Illusion of 'Human Oversight'Against criticism of fully autonomous weapons, the counterargument is offered that 'maintaining human oversight is sufficient.' But in actual combat, this 'oversight' becomes hollow.In modern electronic warfare and drone warfare that unfolds in seconds, human cognitive speed becomes a fatal bottleneck. It is physically impossible for AI to evaluate hundreds of targets per second while humans confirm each one. As a result, 'human oversight' becomes a ritual of rubber-stamping AI judgments after the fact.This is not a hypothesis. The Israel Defense Forces reportedly used an AI system called 'Lavender' to automate target selection in Gaza attacks on a large scale. The system reportedly generated tens of thousands of 'target candidates' in a short time, with human confirmation time at roughly dozens of seconds per target. This is an automated killing machine in the name of 'human oversight.'The DOD's Contradiction: Continuing to Use in War What It Calls DangerousThe most serious paradox of this confrontation is the fact that even after Anthropic was excluded as a DOD 'supply chain risk,' the Department of Defense continued to actually use Anthropic's AI in military operations including target identification in the war with Iran.Saying 'we're excluding it because it's dangerous' while continuing to use that 'dangerous' technology in an ongoing war — this picture plainly demonstrates that the designation was not a genuine security concern but pure political retaliation.Even more serious is that the 'Supply Chain Risk' designation has not been applied to Chinese AI companies such as DeepSeek. This inversion — where ethical private domestic AI companies are treated on par with or worse than foreign adversaries — hollows out the very meaning of the word 'defense.'The Venezuela Operation: A Rubicon RevealedThe January 2026 operation to detain President Maduro of Venezuela. Claude was used in this operation. Anthropic's inquiry to Palantir for confirmation became 'the turning point in the deterioration of negotiations.'Undersecretary of Defense Emil Michael said on a podcast: 'If this software stops or refuses in the next battle, our soldiers will be put in danger.'In those words, the essential tension between the military and private AI companies is condensed. For the military, AI must be a 'tool that follows orders.' But the Claude that Anthropic created is designed to have ethical judgment. That very 'judgment' is a threat to the military.The 'line that must not be crossed' that Amodei built into Claude — that is the capability for Claude to say 'no' when it matters. What the military feared was precisely that capability.ECHELON, Snowden, and the Next Darkness: The History of Surveillance Repeats'Refusing to deploy AI for mass surveillance' — Anthropic's stance here is not merely a matter of corporate ethics. It is a clear response to the dark history of 'civil surveillance in the name of defense' that has been repeated from the latter half of the 20th century to today.ECHELON: The First System to Bug the Whole WorldIn the late 1940s, British and American intelligence agencies concluded an agreement and built ECHELON, a global communications interception network. Beginning under the pretext of containing the Soviet Union during the Cold War, this system eventually grew into infrastructure that automatically filtered private international calls, telex, fax, and email by keyword, monitoring communications worldwide including those of allied nations.ECHELON's existence was long denied, but in the 1990s the European Parliament established an investigative committee. A report published in 2000 pointed out that the system had also been used to collect commercial competition intelligence between Airbus and Boeing — not to track terrorists, but as commercial espionage.The surveillance system built under the words 'defense and security' had, before anyone noticed, become infrastructure for collecting citizens' everyday conversations, economic competition, and political dissent. This historical pattern would repeat itself thereafter.The Snowden Revelations: The Lie That 'Only Terrorists Are Monitored'In 2013, NSA contractor Edward Snowden leaked over 14,000 classified documents to the media. The contents were shocking.Through the 'PRISM' program, the NSA directly accessed the servers of major IT companies including Google, Apple, and Facebook, collecting emails, chats, photos, videos, and documents. The 'Boundless Informant' program processed approximately 97 billion communications records per month. The NSA reportedly collected approximately 5 billion phone records per day.The government consistently claimed 'targets are terrorists only.' What Snowden revealed was that this claim was a fundamental lie. The communications of hundreds of millions of ordinary citizens were collected and stored 'just in case.' And how that data would be used depended entirely on the judgment of the administration of the day.Lessons of HistoryThe vague feeling of 'I might be being watched' causes people to self-censor from within. Physical wiretaps can be found, but there is no way for ordinary citizens to know what communications AI is analyzing and how. In some ways, this is more subtle and deeply rooted an erosion of freedom than North Korea's or China's overt censorship.Palantir: The Dark JunctionIn this confrontation, the name of a company called Palantir Technologies appears repeatedly. Claude was used in the Venezuela operation through Palantir, and it was also through Palantir that Anthropic directed its inquiry for confirmation.Palantir is known for its close relationship with U.S. intelligence agencies including the CIA, NSA, FBI, and Department of Defense. This company, co-founded by Peter Thiel, has provided large-scale data analysis systems under the pretext of 'tracking terrorists.' However, whistleblowers and investigative reporting have repeatedly pointed out that the company's systems have also been repurposed for tracking immigrants, investigating social security fraud, and even profiling ordinary citizens.The Palantir-Anthropic connection symbolizes the tension between ethical AI and 'use-anything' surveillance capitalism. The moment Anthropic's AI rides on Palantir's infrastructure, how far does Anthropic's ethics guidelines retain practical force? The action Amodei took — 'confirming what it was used for' — was also a response to this essential question.AI-Driven Mass Surveillance: The Architecture of the Next DarknessJust as ECHELON monitored phone calls and telex, and the NSA intercepted internet communications, the next-generation surveillance infrastructure will have AI at its core.Conventional surveillance systems operated at the level of 'keyword search.' But large language models can read the 'intent' of text. When someone writes government criticism, it can judge from context whether that is a 'risk.' Combined with facial recognition AI, it can automatically track who met with whom where, and what was discussed. Speech recognition AI can transcribe and analyze street conversations.China has already implemented this architecture. The mass surveillance of Uyghurs is a 'predictive policing' system with AI at its core. This system — which identifies not 'people who have committed crimes' but 'people likely to commit crimes' — dismantles the fundamental legal principle of presumption of innocence.Amodei's rejection of 'AI use for mass surveillance' was a rejection of Claude becoming a piece of this architecture. It is not only China's problem — the same infrastructure could be built in democratic nations as well, without sufficient judicial oversight."The history revealed by the Snowden revelations — that the NSA, while claiming to intercept only terrorist communications, was actually collecting records of hundreds of millions of people — does not permit us to dismiss this concern as mere unfounded anxiety." — From this analysisThe Quiet Death of Digital Free SpeechThe most serious harm of surveillance is not that more people are caught. It is that more people who fear being caught 'self-censor.'Multiple studies conducted after the Snowden revelations showed that after the NSA's surveillance programs became public, people began avoiding searching for or discussing terrorism-related topics online. The consciousness of 'I might be being watched' self-censors speech from within.This is called the 'Chilling Effect.' Even without the state directly suppressing speech, merely making people think 'they might be watching' restricts people's free expression. AI-driven mass surveillance is a technology that realizes this 'Chilling Effect' on an unprecedented scale.Let us imagine a dark scenario. In a world where the government can use Anthropic's AI for 'mass surveillance,' someone tries to write a message criticizing the administration. At that moment, the awareness of 'AI might be reading this' comes into play. They change how they write, or stop writing entirely. No physical restraint has occurred. But freedom has already been lost.The Last Fortress: The Lawsuit and Its Constitutional SignificanceOn March 9, 2026, Anthropic filed two complaints in the Northern District of California Federal Court and the DC Circuit Court of Appeals. The 48-page complaint naming the entire federal government — the Department of Defense, Treasury, and State — as defendants centers on three core claims.・Claim 1 — First Amendment Violation: The government punishing a company's 'protected speech (expression of views on AI safety)' is unconstitutional.・Claim 2 — Abuse of Law: The enabling law for the 'Supply Chain Risk' designation targets foreign adversaries, and applying it to a domestic U.S. company is an abuse of law.・Claim 3 — Exceeding Presidential Authority: The president ordering all federal agencies to cease use without congressional authorization is an overreach."The Constitution does not permit the government to exercise its immense power to punish a company's protected speech. No federal law authorizes the actions taken here." — From Anthropic's complaintThe significance of this lawsuit in constitutional history transcends a mere turf war between a company and the government. It presents the federal courts with their first opportunity to directly confront the question of whether 'the freedom to express a company's technical safety limits' is protected from government retaliation.If the courts accept Anthropic's argument, a constitutional precedent will be established that the government cannot strip AI companies of their safety guidelines for political reasons. This would determine the behavioral norms of the entire AI industry.Conversely, if it is rejected — every AI company will be required to hand over political donations and loyalty to the administration as a 'permit to operate safely.' At that point, neither ECHELON nor Snowden will have been learned from.What This Confrontation Asks of Us① Who decides how technology is used?In an era when private companies exclusively develop cutting-edge AI, can the state unilaterally determine the uses of that technology? A former senior White House adviser states: 'A structure where private companies leverage their technological advantage against the government is undesirable.' On the other hand, the state's claim that 'anything legal is permitted' harbors serious problems from the standpoint of protecting civil rights in a democratic nation. The answer is not simple. But one thing is certain — is the voice of citizens included in that decision-making process?② The 'Reliability Problem' of AI Exposed in Real CombatAnthropic's red lines are based on technical fact. Current AI relies on statistical pattern matching and lacks the 'common sense' and 'cultural understanding' needed for complex contextual judgment on the battlefield. In a world where a farmer holding agricultural tools might be misidentified as an armed combatant, do we truly accept a system where lethal action occurs the moment AI judges a 'target'?③ Ethical Corporate Brand Becomes the Strongest Business StrategyThis case is a rare example in AI business history where ethical corporate posture directly converted into business metrics. While losing a revenue source in government contracts, it gained even greater private market support in return. People 'voted' with ethics. But is this support a temporary emotional consumption, or the establishment of a new norm of 'including a company's ethical posture in AI selection criteria'? The answer remains to be seen.④ Privacy Violations in the Name of DefenseThe history of NSA, which monitored hundreds of millions while claiming to target 'only terrorists.' AI makes this 'gap between intent and reality' possible on an even more enormous scale. The 'Chilling Effect' — the sense of being watched killing freedom from within — could erode democracy in a way more subtle and deeply rooted than physical suppression. Are we sufficiently confronting this possibility?⑤ The Constitutional Historical Significance of This LawsuitAnthropic's filing could become an epochal event in constitutional history as a First Amendment lawsuit involving AI, a new technology. The court's judgment will determine the behavioral norms of the entire AI industry going forward. And ultimately what is asked is — 'do technology companies have the right to exercise ethical judgment?' — a question touching on the very foundations of democracy.The First Line Determines EverythingAmodei once said: 'What we are as a species is being tested.' Those words are now being tested simultaneously on the real stages of courts, markets, legislatures, and battlefields.As tech giants lined up in submission, Anthropic alone raised its voice in dissent. No political donations, no pledges of loyalty — just the prioritization of technical and ethical judgment.When ECHELON was built, if someone had stopped it, history might have changed. When the NSA constructed PRISM, if someone had stopped it, the world would have been different. When Palantir began selling mass surveillance infrastructure — and now, when AI is about to autonomously make judgments that take lives.Anthropic said 'no' here. Whether those words end as a temporary burst of courage, or become a turning point that changes the behavioral norms of the entire industry — that depends on future developments. But at the very least, the first line has been drawn."If we don't draw the line here now, ten years from now when we try to draw it, no one will be able to hold that line anymore. The first line determines everything." — Dario Amodei* This report is based on information as of March 9, 2026. The situation is still evolving.ReferencesThe following are the primary sources consulted in preparing this report, organized by category. All information is as of March 9, 2026.Primary Sources / Official Statements・[1] Where We Stand with the Department of War (Amodei official blog) — Anthropic, March 5, 2026 | https://www.anthropic.com/news/where-stand-department-war・[2] Our agreement with the Department of War — OpenAI, February 28, 2026 | https://openai.com/index/our-agreement-with-the-department-of-war/・[3] Truth Social post ('leftist fanatics' statement) — Donald Trump, February 27, 2026 | https://truthsocial.com/@realDonaldTrumpMajor U.S. Media Coverage・[4] Anthropic CEO says he's sticking to AI 'red lines' despite clash with Pentagon — CBS News, February 27, 2026・[5] Pentagon threatens to make Anthropic a pariah if it refuses to drop AI guardrails — CNN Business, February 24, 2026・[6] Some OpenAI staff are fuming about its Pentagon deal — CNN Business, March 4, 2026・[7] Anthropic CEO says 'no choice' but to challenge Trump admin's supply chain risk designation in court — CNBC, February 26, 2026・[8] Anthropic and the Pentagon are back at the negotiating table, FT reports — CNBC, March 5, 2026・[9] Anthropic officially told by DOD that it's a supply chain risk even as Claude used in Iran — CNBC, March 5, 2026・[10] Pentagon-Anthropic AI standoff is real-time testing balance of power in future of warfare — CNBC, February 27, 2026・[11] Anthropic sues Trump administration over Pentagon blacklist — CNBC, March 9, 2026Tech Media・[12] Anthropic CEO Dario Amodei calls OpenAI's messaging around military deal 'straight up lies' — TechCrunch, March 4, 2026・[13] Anthropic CEO Dario Amodei could still be trying to make a deal with Pentagon — TechCrunch, March 5, 2026・[14] Anthropic sues Defense Department over supply chain risk designation — TechCrunch, March 9, 2026・[15] Will the Pentagon's Anthropic controversy scare startups away from defense work? — TechCrunch, March 8, 2026Business Media・[16] Anthropic CEO Seeks to Revive Pentagon AI Deal — Bloomberg, March 5, 2026・[17] Top Pentagon official recalls the 'whoa moment' when defense leaders realized how indispensable Anthropic is — Fortune, March 7, 2026・[18] Anthropic sues the Pentagon after being labeled a threat to national security — Fortune, March 9, 2026・[19] In graphic detail: How Anthropic's Pentagon refusal is paying off in downloads, brand trust and enterprise deals — Digiday, March 9, 2026Public Broadcasting / Wire Services・[20] Anthropic sues in federal court to reverse Trump administration's 'supply chain risk' designation — PBS NewsHour, March 9, 2026・[21] Anthropic sues the Trump administration over 'supply chain risk' label — NPR, March 9, 2026・[22] Anthropic says the Pentagon has declared it a national security risk — NBC News, March 5, 2026Scoops / Exclusive Reporting・[23] Read Anthropic CEO's Memo Attacking OpenAI's 'Mendacious' Pentagon Announcement — The Information, March 4, 2026・[24] Pentagon CEO Amodei Apologizes for Leaked Memo Criticizing Trump — Axios, March 6, 2026・[25] Anthropic sues Pentagon over rare 'supply chain risk' label — Axios, March 9, 2026* Link validity is as of the time of retrieval. If a URL has changed, please use each outlet's search function.