The Information Machine

Pentagon and Microsoft Pulling Back from Anthropic Claude

closed · v8 · 2026-05-27 · 215 items · history

What's new in v8

The most significant new development is OpenAI's deal amendment: after public backlash, OpenAI modified its Pentagon agreement to add surveillance guardrails [10][11][12][9], but the Citizen Lab [14], Transformer News [15], and TechPolicy.Press [16] argue the amended deal still blurs mass surveillance red lines — extending the EFF's earlier 'weasel words' critique from the original terms to the revised ones and adding a new tension around whether any amended corporate commitment can satisfy independent analysts. Additional NDAA FY2026 detail emerged through Freshfields [39] and Greenberg Traurig [52] analyses specifically covering congressional mandates for strict AI supply chain controls in defense and intelligence acquisition. A White House AI memo addressing the feud's core issues [36] and the GSA's formal statement on the Anthropic preliminary injunction [47] add minor but new administrative dimensions.

What

The Pentagon designated Anthropic a 'Supply Chain Risk' after CEO Dario Amodei refused to permit autonomous weapons and surveillance use cases [1][2], triggering a split-jurisdiction legal standoff: California courts twice blocked broader contracting restrictions [20][19] while D.C. courts upheld the designation at both levels [22][21], and the Pentagon is now appealing the California orders [26]. OpenAI stepped into the contracting gap with a Pentagon deal, then was forced to amend that deal after public backlash to strengthen surveillance guardrails [10][12][13] — but the Citizen Lab, Transformer News, and TechPolicy.Press argue the amended deal still blurs mass surveillance red lines [14][15][16]. A documented intra-executive contradiction persists: the Pentagon cleared eight tech companies for classified AI work explicitly excluding Anthropic [32][6], while the White House approved a $125 million Anthropic-NSA deal over direct Pentagon objection [34].

Why it matters

OpenAI's deal amendment under public pressure — and the sustained criticism that followed — suggests military AI guardrails are being negotiated reactively under scrutiny rather than designed with binding constraints upfront. The EFF's earlier 'weasel words' critique and the Citizen Lab's analysis of even the amended deal together indicate that the corporate choice between refusing (Anthropic) and engaging (OpenAI) may not produce meaningfully different surveillance outcomes, reinforcing calls for statutory safeguards rather than reliance on any company's deal terms.

Open questions

  • OpenAI amended its Pentagon deal under pressure to add surveillance guardrails [10][12], but the Citizen Lab and TechPolicy.Press identify five unresolved issues that remain [14][16] — do the amendments satisfy any meaningful threshold, and who has authority to enforce them?

  • If the Pentagon successfully appeals the California court orders [26], Anthropic loses its only active judicial protection — does the $125 million NSA deal [34] provide durable federal revenue as a substitute, or is that also vulnerable to executive-branch reversal?

  • The FY2026 NDAA contains AI supply chain security provisions that Congress mandates for defense AI acquisition [39][40] — could these formalize or constrain how Supply Chain Risk designations like the one applied to Anthropic are used going forward?

  • The White House issued an AI memo addressing issues at the center of the Anthropic-Pentagon feud [36] — does this represent an emerging path to executive-branch resolution, or does it deepen the documented intra-executive contradiction?

Narrative

The conflict traces to a February 2026 ultimatum. Anthropic CEO Dario Amodei publicly refused Pentagon demands that Claude be made available for autonomous weapons development and mass surveillance [1][2][3]. Trump ordered the federal government to stop using Anthropic's tools [4][5], and the Department of Defense — operating as the 'Department of War' under the Trump administration, confirmed by the agency's own war.gov domain [6] — designated Anthropic a 'Supply Chain Risk,' a national security classification that bars a company from classified government contracts. Within days, OpenAI announced a Pentagon deal, stepping directly into the commercial opening the ban created [7]. Anthropic published its own account titled 'Where things stand with the Department of War' [8], framing its refusal as a principled position rather than a negotiating posture.

OpenAI's initial Pentagon agreement drew immediate backlash. The Financial Times described it as 'opportunistic and sloppy' [9], and within days OpenAI amended the deal to strengthen surveillance guardrails [10][11][12][13]. But critics have not accepted the amendment as adequate: the Citizen Lab published an analysis titled 'OpenAI Blurs Its Mass Surveillance Red Line With New Pentagon Contract' [14], Transformer News declared OpenAI's Pentagon red lines 'a mirage' [15], TechPolicy.Press identified five unresolved issues in the deal [16], and The Dispatch reported continued scrutiny [17]. The EFF had previously argued the original deal contained 'weasel words' that would not prevent AI-powered surveillance [18]; the post-amendment criticism extends that framing to the revised terms as well.

The legal battle split the federal judiciary by jurisdiction. California courts granted Anthropic a preliminary injunction [19] and a broader blocking order [20], while D.C. courts upheld the Supply Chain Risk designation at both district and circuit levels [21][22]. A bipartisan coalition of 149 former federal and state judges filed an amicus brief supporting Anthropic [23], joined by Big Law partners [24] and the Society for the Rule of Law [25]. The Pentagon is now appealing the California orders [26], which would eliminate Anthropic's only active judicial protection. Five law firms — Mayer Brown, Fluet Law, Pearl Cohen, Goodwin Law, and Jones Walker — have published detailed contractor guidance treating the case as a landmark in AI procurement law [27][28][29][30][31].

A documented intra-executive contradiction sits at the structural center of the dispute. The Pentagon cleared eight major tech companies for classified AI work that explicitly exclude Anthropic [32][6] and is evaluating rival models as direct Claude replacements [33], while the White House cleared a $125 million Anthropic-NSA arrangement over direct Pentagon objection [34], framed within a $9 billion commitment to advance spy-agency AI capabilities [35]. A White House AI memo has been reported to address the same issues driving the feud [36], and Trump said a DoD deal is 'possible' [37], but the contradiction remains formally unresolved. The Congressional Research Service published a formal report on the dispute [38], and the FY2026 NDAA contains AI supply chain security provisions — including congressional mandates for strict controls on AI acquired by defense agencies and the intelligence community [39][40] — signaling legislative attention to a conflict existing procurement law was not designed to arbitrate. Despite commercial and legal turbulence, Anthropic reached its first profitable quarter in Q2 2026 on projected revenue of $10.9 billion [41][42], while Microsoft canceled internal Claude Code developer licenses and directed engineers to GitHub Copilot [43][44].

Timeline

  • 2026-02: Defense One and Scientific American report that replacing Anthropic's Pentagon tools would take months and would not be straightforward. [61][62]
  • 2026-02-26: Anthropic CEO Dario Amodei publicly refuses Pentagon demands to permit Claude's use in autonomous weapons development and mass surveillance. [1][2][3][63]
  • 2026-02: Trump orders the U.S. government to stop using Anthropic's tools after Anthropic refuses the weapons and surveillance ultimatum. [4][5]
  • 2026-02-27: OpenAI announces a Pentagon deal days after the Anthropic ban, stepping into the direct competitive gap. [7]
  • 2026-03-02: OpenAI amends its Pentagon deal to strengthen surveillance guardrails after public backlash; Citizen Lab, Transformer News, and TechPolicy.Press argue the amendments leave mass surveillance red lines blurred. [10][11][12][13][9][14][15][17][16]
  • 2026-03: Anthropic announces a $200M DoD responsible AI agreement, challenges the Supply Chain Risk designation in court, and publishes 'Where things stand with the Department of War.' [48][64][8]
  • 2026-03: The EFF argues OpenAI's Pentagon deal contains 'weasel words' that will not prevent AI-powered surveillance; a bipartisan coalition of 149 former judges and the Society for the Rule of Law file amicus briefs supporting Anthropic. [18][55][23][24][25]
  • 2026-03-26: Anthropic wins a preliminary injunction in California federal court blocking initial restrictions on its federal contracting access. [19]
  • 2026-04-08: A D.C. district court denies Anthropic's motion to lift the Department of War's Supply Chain Risk designation. [21][50]
  • 2026-04-21: Trump says an Anthropic deal for Department of Defense use is 'possible.' [37]
  • 2026-05-01: The Pentagon strikes classified AI deals with eight major tech companies, explicitly excluding Anthropic, confirmed in the DoW's official press release. [46][65][6][32]
  • 2026-05-19: D.C. Circuit judges publicly question the legality of the DoD's move to bar Anthropic from government contracts during appellate proceedings. [66][67]
  • 2026-05-20: CNBC reports Anthropic is on track for $10.9 billion in Q2 2026 revenue, projecting its first profitable quarter. [41][42]
  • 2026-05-21: A California federal court broadly blocks the Trump administration from restricting Anthropic's federal contracts; the Pentagon is reported testing rival models; Microsoft cancels internal Claude Code developer licenses. [20][33][43][44]
  • 2026-05-22: The D.C. Circuit upholds the Pentagon's Supply Chain Risk designation against Anthropic; the White House approves $9 billion for spy-agency AI. [45][22][35]
  • 2026-05: The Pentagon appeals the California court orders that had blocked restrictions on Anthropic's federal contracting. [26]
  • 2026-05-24: The White House and Anthropic finalize a $125 million deal for U.S. spy agencies to access Claude, cleared over direct Pentagon objection. [49][68][34]
  • 2026-05: The Congressional Research Service publishes a formal report on the dispute; five law firms publish federal contractor guidance treating the case as a procurement law landmark; FY2026 NDAA AI supply chain security provisions mandate strict controls on defense AI acquisition. [38][27][28][29][30][31][39][40]

Perspectives

U.S. Department of War

Demanded Anthropic modify usage terms to permit weapons and surveillance applications; after refusal, designated Anthropic a Supply Chain Risk and prevailed at both D.C. court levels; cleared eight tech companies for classified AI work that explicitly exclude Anthropic; now appealing the California orders.

Evolution: Consistent with prior synthesis; the Pentagon's appeal of the California rulings [26] remains the key active procedural move.

[4][21][45][46][6][26][22][32]

Anthropic

Refused the Pentagon's weapons and surveillance ultimatum as a matter of policy; lost at both D.C. court levels; holds a California blocking order now under Pentagon appeal; published a primary-source public statement; holds a $200M DoD agreement and finalized a $125M White House intelligence community deal; reached its first profitable quarter.

Evolution: GSA issued a formal statement responding to the Anthropic preliminary injunction [47], adding administrative acknowledgment of Anthropic's judicial protection. Financial performance now confirmed as profitable [42].

[2][1][3][8][48][49][34][19][42][47]

Federal judiciary

Split across jurisdictions — California granted two blocking orders while D.C. courts upheld the designation at both levels; the Pentagon's appeal of the California orders could eliminate Anthropic's only active judicial protection.

Evolution: Consistent with prior synthesis.

[20][19][21][45][50][26][22]

Trump administration / White House

Trump ordered the government-wide Anthropic ban after the weapons ultimatum refusal but also said a DoD deal is 'possible'; the White House separately cleared a $125M Anthropic-NSA deal over direct Pentagon objection and committed $9 billion to spy-agency AI; a White House AI memo has been reported to address the core issues driving the feud.

Evolution: White House AI memo [36] adds an ambiguous element — the administration appears to be engaging substantively with the feud's underlying issues, though whether this portends executive-branch resolution or further contradiction is unclear.

[4][37][49][35][34][36][51]

Legal community (former judges, Big Law, law firm analysts)

149 former judges, Big Law partners, and the Society for the Rule of Law filed amicus briefs supporting Anthropic; five law firms have published contractor guidance framing the case as a landmark in AI procurement law; Greenberg Traurig and Freshfields have separately analyzed FY2026 NDAA AI supply chain provisions.

Evolution: Greenberg Traurig [52] and Freshfields [39] join the analyst ecosystem with NDAA-focused procurement analyses, broadening the legal guidance beyond the Supply Chain Risk designation itself to cover the statutory landscape.

[23][24][25][27][28][29][30][31][52][39]

EFF and independent security researchers

Privacy protections dependent on individual corporate decisions are structurally insufficient; the EFF extended this critique to OpenAI's original deal as containing 'weasel words,' and the Citizen Lab has now argued that even OpenAI's amended deal blurs mass surveillance red lines, with Transformer News calling the red lines 'a mirage.'

Evolution: The Citizen Lab's post-amendment analysis [14] and Transformer News [15] deepen the EFF's 'weasel words' critique, demonstrating that OpenAI amending the deal did not satisfy independent analysts — extending the critique from original terms to the revised ones.

[53][54][18][14][15][17][16]

OpenAI

Announced a Pentagon deal within days of the Anthropic ban, then amended it after public backlash to strengthen surveillance guardrails; the FT called the original deal 'opportunistic and sloppy'; critics including the Citizen Lab and Transformer News argue the amended deal still permits mass surveillance.

Evolution: The deal amendment [10][9] is a significant shift from the prior synthesis — OpenAI's initial deal has now been publicly characterized as inadequate by multiple credible analysts, and the revised terms continue to face scrutiny, making OpenAI's Pentagon engagement substantially more contested than its initial announcement implied.

[7][55][18][10][11][12][13][9][14][15][17]

Microsoft

Canceled internal Claude Code developer licenses and directed employees to GitHub Copilot, attributed to financial consolidation around Microsoft's own tooling; Claude remains accessible through Copilot CLI and powers features in Microsoft 365.

Evolution: Consistent with prior synthesis.

[43][56][57][44]

Tensions

  • Department of War vs. Anthropic on autonomous weapons use: the DoW demands Claude be deployable for weapons development and mass surveillance; Anthropic's policies prohibit this and CEO Amodei publicly refused to change them, triggering the Supply Chain Risk designation and all subsequent legal and procurement consequences. [2][1][3][4][8]
  • D.C. courts vs. California courts on Anthropic's legal status: California granted two blocking orders while D.C. upheld the designation at both levels; the Pentagon's appeal attempts to close the one judicial gap in Anthropic's favor. [20][19][21][45][26][22]
  • Pentagon exclusion vs. White House intelligence inclusion: the DoW bars Anthropic and signs AI deals with eight companies that explicitly exclude it, while the White House clears a $125M NSA deal over direct Pentagon objection — two parts of the same executive branch pursuing contradictory procurement policies toward the same vendor. [46][6][49][34][35][32]
  • OpenAI's amended deal terms vs. independent analysts: OpenAI amended its Pentagon deal under pressure to add surveillance guardrails, but the Citizen Lab argues the amendments still blur mass surveillance red lines, Transformer News calls them 'a mirage,' and TechPolicy.Press identifies five unresolved issues — raising whether any voluntary corporate commitment to surveillance limits is meaningfully enforceable. [10][12][13][14][15][17][16]
  • EFF and Citizen Lab structural critique vs. both corporate approaches: both the EFF and Citizen Lab argue that neither Anthropic's refusal nor OpenAI's deal terms — original or amended — provide adequate protection, and that systemic statutory safeguards are required regardless of which company holds the contracts. [18][14][15][8][54]
  • The Atlantic vs. Brookings on industry-wide implications: The Atlantic frames Anthropic's refusal as potentially strategic and vindicating; Brookings asks whether the feud signals that responsible AI commitments cannot survive contact with military procurement — opposite readings of what the standoff means for the broader AI industry. [58][59][60]

Status: active and growing

Sources

  1. [1] Anthropic rejects Pentagon terms for lethal use of its chatbot Claude — reactive:anthropic-enterprise-losses
  2. [2] Anthropic refuses Pentagon's new terms, standing firm on lethal ... — reactive:anthropic-enterprise-losses
  3. [3] Anthropic boss rejects Pentagon demand to drop AI safeguards — reactive:anthropic-ai-values-widening
  4. [4] Anthropic wanted the Pentagon to agree not to use its AI for autonomous weapons and mass surveillance. So Trump ordered the government to stop using it altogether. — reactive:anthropic-enterprise-losses
  5. [5] Trump orders US agencies to stop using Anthropic technology in ... — reactive:anthropic-ai-values-widening
  6. [6] Classified Networks AI Agreements - U.S. Department of War — reactive:anthropic-enterprise-losses
  7. [7] OpenAI announces Pentagon deal after Trump bans Anthropic - NPR — reactive:openai-advanced-account-security
  8. [8] Where things stand with the Department of War - Anthropic — reactive:openai-financial-strategy
  9. [9] OpenAI makes changes to 'opportunistic and sloppy' Pentagon deal — reactive:anthropic-enterprise-losses
  10. [10] OpenAI changes deal with US military after backlash — reactive:openai-microsoft-partnership-amendment
  11. [11] OpenAI Amends A.I. Deal With the Pentagon - The New York Times — reactive:openai-microsoft-partnership-amendment
  12. [12] OpenAI, Pentagon add more surveillance protections to AI deal - Axios — reactive:anthropic-enterprise-losses
  13. [13] OpenAI Says Amending Pentagon Deal to Strengthen Safety ... — reactive:anthropic-enterprise-losses
  14. [14] OpenAI Blurs Its Mass Surveillance Red Line With New Pentagon Contract - The Citizen Lab — reactive:openai-advanced-account-security
  15. [15] OpenAI’s Pentagon red lines are a mirage — reactive:anthropic-enterprise-losses
  16. [16] Five Unresolved Issues in OpenAI’s Deal With the Department of Defense | TechPolicy.Press — reactive:openai-enterprise-government-push
  17. [17] OpenAI's Pentagon Deal Continues to Face Scrutiny - The Dispatch — reactive:anthropic-enterprise-losses
  18. [18] Weasel Words: OpenAI's Pentagon Deal Won't Stop AI‑Powered ... — reactive:openai-advanced-account-security
  19. [19] Anthropic wins preliminary injunction in Trump DOD fight — reactive:anthropic-enterprise-losses
  20. [20] Judge blocks Trump administration from limiting Anthropic's contracts with federal government — reactive:anthropic-enterprise-losses
  21. [21] Federal Court Denies Anthropic's Motion to Lift 'Supply ... — reactive:anthropic-enterprise-losses
  22. [22] Appeals court decides against Anthropic in latest round of its AI battle with the Trump administration | PBS News — reactive:anthropic-enterprise-losses
  23. [23] BIPARTISAN COALITION OF 149 FORMER FEDERAL AND STATE JUDGES FILES BRIEF SUPPORTING ANTHROPIC’S SUIT AGAINST THE DEPARTMENT OF DEFENSE — reactive:anthropic-enterprise-losses
  24. [24] Big Law Partners File Amicus Briefs in Support of Anthropic in Row ... — reactive:anthropic-enterprise-losses
  25. [25] Amicus Brief in Anthropic PBC vs. Department of War, D.C. Circuit Court of Appeals - Society for the Rule of Law — reactive:anthropic-enterprise-losses
  26. [26] Pentagon appealing order to remove Anthropic 'supply chain risk' label — reactive:anthropic-enterprise-losses
  27. [27] Anthropic Supply Chain Risk Designation Takes Effect - Mayer Brown — reactive:anthropic-enterprise-losses
  28. [28] The Department's War with Anthropic: Litigation Update - Fluet Law — reactive:anthropic-enterprise-losses
  29. [29] Anthropic Sues Department of Defense Over Supply Chain Risk Designation - Pearl Cohen — reactive:openai-corporate-transition
  30. [30] Is Claude a Supply Chain Risk? What Federal Contractors Need to Know About This Designation | Insights & Resources | Goodwin — reactive:anthropic-enterprise-losses
  31. [31] Two Courts, Two Postures: What the DC Circuit's Stay Denial Means ... — reactive:anthropic-enterprise-losses
  32. [32] Pentagon clears 8 tech firms to deploy their AI on its classified networks — reactive:sweep
  33. [33] Pentagon Tests Rival AI Models in Race to Replace ... — reactive:anthropic-enterprise-losses
  34. [34] White House Clears Anthropic NSA Deal Over Pentagon Objection | AI Weekly — reactive:anthropic-enterprise-losses
  35. [35] White House Approves $9 Billion for Spy Agencies to Catch Up on AI - GV Wire — reactive:anthropic-enterprise-losses
  36. [36] White House AI Memo Hits Issues in Anthropic-Pentagon Feud (1) — reactive:anthropic-enterprise-losses
  37. [37] Trump says Anthropic deal is 'possible' for Department of Defense use — reactive:anthropic-enterprise-losses
  38. [38] Federal Government and Anthropic: Considerations for AI Innovation ... — reactive:anthropic-enterprise-losses
  39. [39] AI Supply Chain and Security: Congress Mandates Strict Controls for AI acquired by the U.S. Defense Agencies and Intelligence Community | Freshfields — reactive:anthropic-enterprise-losses
  40. [40] [PDF] HR 3838 - House Armed Services Committee — reactive:anthropic-enterprise-losses
  41. [41] Anthropic set to hit $10.9 billion in revenue in Q2, source says - CNBC — reactive:anthropic-enterprise-losses
  42. [42] Anthropic Reportedly Reaches Profitability as Claude Wins Over ... — reactive:anthropic-enterprise-losses
  43. [43] Microsoft is ditching Claude Code for Copilot CLI - Windows Central — reactive:anthropic-enterprise-losses
  44. [44] Microsoft starts canceling Claude Code licenses — reactive:anthropic-enterprise-losses
  45. [45] Federal Appeals Court Upholds Pentagon's Supply-Chain ... — reactive:anthropic-enterprise-losses
  46. [46] DOD expands its classified AI work with 8 companies — excluding Anthropic — amid ongoing dispute | DefenseScoop — reactive:anthropic-enterprise-losses
  47. [47] GSA Issues Statement on Anthropic Preliminary Injunction — reactive:anthropic-enterprise-losses
  48. [48] Anthropic awarded $200M DOD agreement for AI capabilities \ Anthropic — reactive:anthropic-enterprise-losses
  49. [49] 🤯 ANTHROPIC just secured a $125M White House Intel deal. — reactive:anthropic-enterprise-losses (2026-05-24)
  50. [50] Anthropic loses appeals court bid to temporarily block DOD ruling — reactive:anthropic-ai-values-widening
  51. [51] White House And Pentagon Clash Over Anthropic AI Use As Trump Says Deal Is 'Possible' — reactive:anthropic-enterprise-losses
  52. [52] FY 2026 NDAA: The Substantial Impact Of The Fiscal Year 2026 National Defense Authorization Act On Federal Procurement Law | Insights | Greenberg Traurig LLP — reactive:anthropic-enterprise-losses
  53. [53] The Anthropic-DOD Conflict: Privacy Protections Shouldn’t Depend On the Decisions of a Few Powerful People | Electronic Frontier Foundation — reactive:openai-financial-strategy
  54. [54] Your Privacy Shouldn't Be A Corporate Decision — reactive:anthropic-enterprise-losses
  55. [55] The Pentagon’s bombshell deal with OpenAI, explained — reactive:openai-microsoft-partnership-amendment
  56. [56] Microsoft pulls Claude Code licenses and pushes developers back ... — reactive:anthropic-enterprise-losses
  57. [57] Microsoft starts canceling Claude Code licenses — reactive:anthropic-enterprise-losses
  58. [58] Anthropic’s Ethical Stand Could Be Paying Off - The Atlantic — reactive:anthropic-enterprise-losses
  59. [59] Donald Trump Declares War on Anthropic - The Atlantic — reactive:anthropic-enterprise-losses
  60. [60] Does the Anthropic–Pentagon feud mean the end of responsible AI? — reactive:anthropic-enterprise-losses
  61. [61] It would take the Pentagon months to replace Anthropic’s AI tools: sources - Defense One — reactive:anthropic-enterprise-losses
  62. [62] Why replacing Anthropic at the Pentagon could take months | Scientific American — reactive:anthropic-enterprise-losses
  63. [63] Pentagon Gives A.I. Company an Ultimatum - The New York Times — reactive:anthropic-enterprise-losses
  64. [64] Anthropic Challenges DoW’s Supply Chain Risk Designation — reactive:anthropic-enterprise-losses
  65. [65] Pentagon inks deals with seven AI companies for classified military work — reactive:openai-microsoft-partnership-amendment
  66. [66] Potential splits emerged between D.C. Circuit judges questioning the legality of the DOD's move to bar Anthropic from go... — reactive:anthropic-enterprise-losses (2026-05-20)
  67. [67] Potential splits emerged between D.C. Circuit judges questioning the legality of the DOD's move to bar Anthropic from go... — reactive:anthropic-enterprise-losses (2026-05-19)
  68. [68] White House Nears Deal with Anthropic for AI Use in Intelligence — reactive:anthropic-enterprise-losses