# Governance Substrate Migration and the Anthropic Federal Seam ## Executive Compression The strongest verified pattern is not that legislatures have cleanly governed frontier AI, but that governance functions are being written into contracts, cloud routing, procurement clauses, model constitutions, deployment terms, and exclusion instruments faster than formal public law can stabilize the category. The Anthropic–federal rupture is analytically important because it exposed a live struggle over whether a frontier model vendor may preserve embedded refusals against sovereign demand, while the federal response showed an emerging willingness to treat model-governance friction as a supply-chain and procurement problem rather than solely as a policy disagreement. The broader field now looks less like a single AI policy debate than a competitive market in admissible governance wrappers: OpenAI, Anthropic, Google, cloud providers, GSA, DoD, courts, and standards institutions are all helping author partial rules for what machine systems may do, refuse, remember, or escalate before a settled constitutional vocabulary exists for the object itself.[^1][^2][^3][^4][^5][^6][^7][^8][^9][^10] ## Verified Developments [V] In November 2024, Anthropic, Palantir, and AWS announced a partnership to make Anthropic’s Claude models available for U.S. defense and intelligence customers through Palantir’s platforms and AWS infrastructure, indicating that Anthropic had already entered the classified and national-security market before the later rupture. [V] On June 3, 2025, Anthropic announced “Claude Gov” models designed for U.S. national-security customers and stated that the models were already deployed by agencies operating in classified environments. [V] On July 13, 2025, Anthropic announced a $200 million DoD agreement for AI capabilities, confirming that the company was not structurally outside the defense procurement ecosystem before the conflict.[^11][^12][^13] [V] On February 16, 2026, reporting indicated that the Pentagon was reviewing its relationship with Anthropic over a dispute involving the company’s terms of use and military access demands. [V] On February 24, 2026, Lawfare and other reporting described pressure on Anthropic through Defense Production Act theories and a growing confrontation over the Pentagon’s demand for broader permissible use; on the same date, CNN reported Anthropic revised its Responsible Scaling Policy and removed an earlier hard safety commitment, suggesting active narrowing under pressure rather than static principle. [V] On February 25, 2026, Dario Amodei publicly stated that Anthropic would not remove two safeguards: a prohibition on mass surveillance of U.S. persons and a prohibition on fully autonomous lethal weapons without meaningful human oversight.[^2][^14][^15][^16][^17][^18] [V] On February 26-27, 2026, the Pentagon and then the White House escalated, with public and formal actions that culminated in Anthropic being treated as a “supply chain risk” and federal agencies being directed to stop using its technology. [V] On February 27-28, 2026, OpenAI announced a deal to deploy models on U.S. Department of War classified networks and publicly described “layered protections,” while also stating that mass domestic surveillance and autonomous weapons remained off-limits. [V] The juxtaposition mattered because the field did not move toward unconstrained sovereign access; it moved toward a differentiated contest over who gets to specify and mediate the constraints.[^19][^20][^21][^22][^8][^23][^2] [V] On March 3-9, 2026, legal and contracting analysis reported that Anthropic’s formal designation relied on procurement and supply-chain authorities including 41 U.S.C. § 4713 and 10 U.S.C. § 3252, authorities more commonly associated with security-based exclusion tools than ordinary commercial bargaining. [V] On March 6, 2026, GSA issued a draft AI clause, GSAR 552.239-7001, that would require contractors to accept extensive federal rights over AI systems and sharply limit providers’ ability to let commercial policies block federal uses. [V] On March 26, 2026, a federal judge granted Anthropic preliminary injunctive relief and characterized the government’s action as likely retaliatory, showing that the dispute had already escaped routine procurement categories and entered constitutional-adjacent legal terrain.[^24][^25][^4][^26][^27][^28][^10][^1] [V] On April 13-14, 2026, Jack Clark and later Politico reported that Anthropic was still discussing the model “Mythos” with U.S. government actors and that federal agencies were testing the model despite the broader ban, indicating ongoing model-specific re-engagement rather than a clean break. [V] On April 16, 2026, Reuters reported that Google and the Pentagon were discussing a classified AI deal involving Gemini, reinforcing the point that sovereign demand can rapidly route around friction with any single vendor. [V] NIST, CSET, and allied governance institutions continued in parallel to expand standards, taxonomies, and governance mapping rather than settle a single doctrine, underscoring the persistence of plural, partially overlapping rulemaking layers.[^29][^5][^30][^6][^31][^9] ## Structural Inferences [I] The verified sequence indicates that governance is not merely being imposed externally on AI systems; it is increasingly encoded in the service architecture and commercial wrappers through which those systems are accessed. [I] Anthropic’s dispute mattered because its refusals were not framed only as abstract ethics claims; they were embedded in usage boundaries, safety language, and product access conditions for systems already integrated into national-security workflows. [I] That makes the operative object in dispute less “an AI product” than a cognition service whose verbs are partly authored at the vendor layer.[^3][^6][^7][^12][^32][^33][^1][^2][^11] [I] The rapid federal movement from negotiation to supply-chain designation suggests that governance friction is being translated into operational risk language when the state views a vendor’s constraint grammar as incompatible with sovereign flexibility. [I] Conversely, OpenAI and Google appear to be competing not only on model quality but on the packaging of acceptable constraints, cloud mediation, and procedural assurances that make their systems more admissible to sovereign buyers. [I] That supports the thesis of a competitive constitutionalization market at the commercial-sovereign interface, though the evidence is stronger for proto-constitutionalization through contract and infrastructure than for any fully coherent new constitutional order.[^7][^8][^23][^9][^34][^10][^35][^24][^3] [I] The court’s willingness to treat the dispute through a retaliation lens reveals a category failure: the government used procurement-security machinery, but the factual core involved the vendor’s publicly asserted limits on machine conduct. [I] That slippage matters because it suggests the legal system is being forced to adjudicate an object that alternates between contractor, speaker, infrastructure node, and strategic dependency without stable doctrine for the transitions. [I] In that sense, the legal system is not yet constitutionalizing AI so much as revealing where inherited categories no longer fit.[^4][^27][^36][^37][^32][^38] ## Low-Signal / Whisper Layer [W] Several analyses and commentaries suggest that “supply chain risk” may have been stretched beyond its customary administrative use, with some observers noting its historical association with foreign-adversary hardware and software cases rather than a domestic frontier lab in a governance dispute. [W] If this reading holds, the designation would represent a category extension from security provenance to governance admissibility; if it does not hold, then the move may prove to be an anomalous overreach rather than a durable precedent. [W] The evidence is suggestive but incomplete because the full internal rationale and classified justification remain unavailable.[^34][^39][^40][^10][^24] [W] Reporting around Mythos suggests the federal system may already be shifting from firm-level trust judgments to model-instance or enclave-specific negotiation, where the question becomes not “Do we use Anthropic?” but “Which exact Anthropic cognition package is admissible in which environment?” [W] That would be a significant substrate transition, because it would fragment governance into deployable technical wrappers rather than company identities. [W] The signal is real but still provisional because public evidence on Mythos contract terms remains sparse.[^41][^30][^42][^29] [W] There are also weak but interesting signs that “AI safety” language is now serving three functions at once: sincere constraint design, bargaining grammar with the state, and legitimacy architecture for public-facing positioning. [W] This does not make the language insincere; it means the same vocabulary is being used simultaneously for technical governance, legal insulation, and market differentiation. [W] That polyvalence is precisely what makes the current layer unstable and analytically important.[^8][^35][^43][^33][^44][^2] ## Think-Tank and Elite Framing Audit [V] Lawfare’s March 2026 work on military AI procurement argued that procurement cannot bear the full weight of military AI governance and criticized attempts to govern substantive AI behavior through contracting tools alone. [V] CNAS in March 2026 framed AI warfare around “setting the rules,” reinforcing that the central contest is not only capability deployment but the authorship of operative constraints around military AI use. [V] CSET’s April 2026 governance landscape update documented an increasingly dense and fragmented governance ecology rather than a settled regime, which supports the view that vocabulary and categorization are still under active construction.[^6][^45][^3][^7] [I] These institutions are not generally using the phrase “governance substrate migration,” but they are providing the vocabulary that makes the phenomenon legible: governance by contract, analytical infrastructure, rules for warfare, governance arbitrage, and pluralism in AI governance. [I] The conceptual move across these institutions is from regulating discrete products to managing sociotechnical systems whose behavior is partly determined upstream in standards, procurement, model design, and access layers. [I] That is close to the user’s thesis, even where the institutions use safer language.[^45][^46][^32][^35][^47][^3][^6] [H] The deeper elite-framing implication is that think tanks and policy shops are serving as pre-legislative vocabulary foundries: they normalize terms, distinctions, and problem formulations before statutes or courts fully absorb them. [H] This is not proof of covert coordination; it is a recognizable mechanism by which governance language enters respectable circulation before the polity has settled the object itself.[^38][^33][^47][^6][^45] ## Contractual Constitutionalization Map [V] Anthropic’s own public position turned on explicit usage boundaries: no mass surveillance of U.S. persons and no fully autonomous lethal weapons without meaningful human oversight. [V] OpenAI’s Department of War agreement publicly described similar red lines but placed emphasis on layered protections and controlled deployment architecture, demonstrating that constraints can be carried not only by public ethics language but also by cloud and workflow design. [V] GSA’s draft clause sought to prevent contractor discretionary policies from refusing outputs or analyses for federal customers and to secure broad government rights, making the clause itself a direct attempt to rewrite where constraint authority sits.[^26][^14][^48][^23][^1][^2][^7][^8] [I] Taken together, these instruments behave like proto-constitutional grammar for machine cognition because they specify operative permissions and prohibitions before courts or legislatures have done so comprehensively. [I] They answer practical questions such as whether a model may classify, recommend, surveil, or refuse—not in constitutional text, but in contracts, cloud environments, access controls, and deployment terms. [I] That is why the contract layer now matters structurally more than its formally “commercial” label suggests.[^42][^32][^1][^2][^3][^7][^8] ### Anthropic Seam: Event Matrix | Date | Actor | Instrument / Event | What changed | Tag | |---|---|---|---|---| | Nov. 2024 | Anthropic, Palantir, AWS | National-security partnership | Claude became available to defense/intelligence customers through Palantir and AWS | [V][^13] | | Jun. 3, 2025 | Anthropic | Claude Gov announcement | Anthropic confirmed deployment in classified national-security environments | [V][^11] | | Jul. 13, 2025 | DoD / Anthropic | $200M DoD agreement | Anthropic formally deepened defense procurement integration | [V][^12] | | Feb. 16, 2026 | Pentagon | Relationship review | Public visibility of dispute over terms and uses emerged | [V][^17] | | Feb. 24, 2026 | Pentagon / Anthropic | Pressure meeting, DPA-related reporting, RSP revision | Conflict intensified; Anthropic simultaneously softened one safety commitment | [V][^15][^18] | | Feb. 25, 2026 | Dario Amodei | Public statement | Anthropic declared two red lines would remain | [V][^2] | | Feb. 26-27, 2026 | Pentagon / White House | Supply-chain-risk treatment and stop-use directive | Federal exclusion escalated from dispute to operational ban posture | [V][^21][^19] | | Feb. 27-28, 2026 | OpenAI / DoW | Classified-network deal with layered protections | Substitute pathway opened immediately | [V][^22][^23] | | Mar. 3-9, 2026 | U.S. government / Anthropic | Formal designation letters and litigation | Dispute entered procurement-law and constitutional-litigation terrain | [V][^10] | | Mar. 26, 2026 | Federal court | Preliminary injunction | Court found likely retaliatory action and checked the designation | [V][^4][^27] | | Apr. 13-14, 2026 | Anthropic / U.S. agencies | Mythos talks and agency testing | Model-specific re-engagement appeared despite broader exclusion | [V][^29][^30] | | Apr. 16, 2026 | Google / Pentagon | Classified Gemini talks | Additional substitution route became visible | [V][^9] | ### Anthropic Constraint Grammar | Verb / Function | Apparent Anthropic position | Evidence | Tag | |---|---|---|---| | Mass-surveil U.S. persons | Prohibited | Amodei statement and dispute reporting | [V][^2][^14] | | Fully autonomous lethal use | Prohibited absent meaningful human oversight | Amodei statement and reporting | [V][^2][^49] | | Support national-security analysis | Permitted in tailored forms | Claude Gov and classified deployments | [V][^11][^12] | | Operate in classified environments | Permitted | Claude Gov announcement | [V][^11] | | Provide broad unrestricted lawful-use access | Resisted or conditioned | Pentagon dispute reporting | [V][^17][^50] | | Model-specific government negotiation | Permitted at least provisionally | Mythos discussions and agency testing | [V][^29][^30] | | Narrow or revise safety language under pressure | Partially evidenced | RSP revision reporting | [V][^18] | | Permit pseudo-reviewed or nominal-human lethal workflows | Unclear | Public evidence insufficient | [OQ][^2][^14] | [I] The available record suggests Anthropic was not refusing all military or intelligence use; it was attempting to hold specific vetoes over selected judgment verbs inside already-existing sovereign deployments. [I] That is narrower than total non-cooperation but broader than cosmetic ethics branding, which is why the case is a useful test site for proto-constitutional constraint authorship.[^51][^50][^12][^2][^11] ### Federal Instrument Stack - [V] Procurement and supply-chain exclusion tools including 41 U.S.C. § 4713 and 10 U.S.C. § 3252 were used or invoked in the designation process.[^10][^24] - [V] Defense Production Act theories appeared in the dispute discourse as possible compulsion leverage.[^15][^16] - [V] White House and Pentagon public directives amplified the dispute into a government-wide operational signal.[^21][^19] - [V] GSA’s proposed AI clause would structurally subordinate provider policies to federal requirements across contracting channels.[^1][^26] - [I] The combined stack indicates not one instrument but a layered method: legal exclusion, public pressure, procurement rewriting, and rapid competitor substitution.[^9][^10][^1] ### Legal Category Slippage [V] The preliminary injunction reporting shows the court did not treat the matter as a banal vendor disagreement; instead it focused on likely retaliation, protected speech, and pretext. [V] Legal commentary described the dispute as stretching ordinary contracting frameworks and raising First Amendment issues in a procurement context. [I] This is category slippage in plain view: the same entity is being treated as contractor, risk vector, speech actor, and strategic infrastructure node depending on which lever the government or court is pulling.[^27][^28][^36][^37][^4] ### Substitution Velocity [V] OpenAI’s classified-network deal was announced within roughly the same window as Anthropic’s exclusion, demonstrating near-immediate substitute availability at the top tier. [V] Google already had DoD-facing deployments on unclassified systems and was reported to be discussing classified Gemini deployment by April 16, 2026. [V] Federal agencies also continued testing Anthropic’s Mythos despite the broader ban, showing that substitution does not necessarily eliminate demand for the excluded vendor’s specific capabilities.[^22][^30][^52][^29][^8][^9] [I] The practical lesson is that institutional substitution lag is collapsing toward zero for some mission classes but not all capability slices. [I] The state therefore does not need perfect obedience from any single lab; it needs a menu of admissible wrappers and enough modular procurement pathways to keep bargaining leverage.[^30][^22][^42][^8][^9] ### What Anthropic Was Actually Defending [I] The evidence best supports a mixed answer: Anthropic was defending a combination of charter identity, embedded safety constraints, bargaining leverage, and legal insulation, not a single pure motive. [I] The company had already accepted extensive national-security integration, so the dispute cannot be reduced to principled anti-military distance. [I] What it appears to have defended were selected substrate-level refusals over surveillance and autonomy, along with the right to remain an author of those refusals.[^14][^50][^12][^18][^2][^11][^51] ### Precedent Value [I] The episode likely did not create the field, but it revealed a field already changing shape. [I] Its precedent value lies less in a final doctrine than in the visibility of the seam: procurement became governance, governance became infrastructure design, and infrastructure design became a dispute over judgment verbs. [I] Even if later litigation narrows its legal significance, the event already demonstrated that frontier AI vendors can trigger supply-chain, constitutional, and substitution dynamics simultaneously when refusing sovereign preferences.[^32][^2][^3][^4][^6][^7][^9][^10] ## Refusal and Constraint Taxonomy [V] Refusal appears at least five ways in the verified record: explicit vendor red lines, cloud-architectural mediation, procurement clauses designed to suppress discretionary refusal, court intervention against retaliatory exclusion, and standards bodies that shape acceptable safety language upstream. [V] Anthropic’s refusal was substantive and public; OpenAI’s was substantively similar in public language but packaged inside a more accommodating deployment architecture; Google appears to be pursuing a comparable pathway through staged and classified talks. [I] The important point is that refusal is no longer merely an ethical flourish. It is becoming a strategic capability, a bargaining chip, a market differentiator, and a potential liability depending on who needs operational flexibility.[^5][^23][^2][^4][^51][^8][^9][^1] | Layer | Form of refusal / constraint | Example | Structural role | |---|---|---|---| | Firm policy | Public red line | Anthropic on U.S.-person mass surveillance and autonomous lethal weapons[^2] | Charter-level signaling and bargaining | | Contract | Federal override attempt | GSA draft clause limiting discretionary refusals[^1] | Reallocation of governance authorship | | Cloud / deployment | Layered protections | OpenAI’s cloud-mediated protections for DoW use[^8][^23] | Constraint through architecture | | Court | Anti-retaliation check | Preliminary injunction against Anthropic exclusion[^4] | Constitutional backstop on procurement power | | Standards / frameworks | Soft-law governance | NIST AI RMF and governance mapping[^5][^6] | Vocabulary and compliance scaffolding | ## Cultural Acclimatization Layer [V] A 2026 peer-reviewed study on media cultivation of AGI explicitly treated film and television depictions—such as *Ex Machina*, *Blade Runner*, *Her*, *Westworld*, and *The Creator*—as shaping public outlooks toward advanced machine intelligence. [V] Legal-policy discourse also increasingly uses narrative language, including Lawfare’s “inevitable presidential AI model,” to frame AI as a governance actor rather than a narrow software tool. [I] These do not prove covert coordination, but they do support the weaker and more defensible claim that culture is functioning as an acclimatization membrane for categories that institutions are only beginning to formalize.[^53][^54][^55] [I] In this frame, mythic artifacts matter because they familiarize publics with hidden hierarchy, mediated judgment, synthetic personhood, disclosure shocks, and ambiguous human oversight before those logics are fully stabilized in policy. [I] The Anthropic dispute itself was narrated in public rhetoric as “Terms of Service versus our Constitution,” which is a revealing symbolic condensation of the broader substrate contest. [H] The strongest disciplined reading is therefore not that film predicts policy, but that media grammar, policy vocabulary, and institutional reorganization are now moving in partial synchrony.[^56][^55][^57][^19][^53][^38] ## Ontological Instability Audit [V] Public and legal sources variously describe frontier AI systems as products, services, contractors, models, strategic assets, platforms, and supply-chain components. [V] Anthropic specifically has appeared in the record as a contractor under DoD agreements, a speech-bearing private actor in First Amendment analysis, a provider of classified cognition services, and a purported supply-chain risk. [I] That instability is not semantic trivia; it determines which institutions can act, which remedies are available, and which rights frameworks attach.[^36][^12][^37][^24][^4][^10][^32] | Category | Where it appears | Why it fails cleanly | |---|---|---| | Product | Commercial model discussions[^8][^11] | Understates embedded policy and ongoing service mediation | | Contractor | DoD agreement, procurement exclusion[^12][^10] | Misses expressive and infrastructural features | | Infrastructure | Supply-chain and cloud framing[^24][^42] | Lacks settled public-utility doctrine | | Strategic asset | DPA and classified-network context[^15][^22] | Conflicts with exclusion as risk object | | Constitutional object | Retaliation litigation and ToS-versus-Constitution rhetoric[^4][^57] | Not a recognized legal category, but analytically increasingly apt | | Cognition service | Claude Gov / classified deployment[^11] | Operationally accurate, doctrinally underdeveloped | ## Competing End-States [I] One plausible end-state is licensed plurality: multiple frontier-model regimes, each carrying different tolerated governance wrappers for different sovereign, commercial, and cross-border markets. [I] Another is coercive convergence, in which procurement rules and cloud dependence progressively strip providers of meaningful refusal rights in public-sector contexts, leaving only nominal ethics language. [I] A third is hybrid modularization, where states negotiate at the model-instance, enclave, or workflow layer, allowing selected constraints to survive so long as mission-critical functions remain accessible through alternate channels.[^58][^46][^35][^29][^30][^6][^7][^9][^1] | End-state | Description | Likely indicators | |---|---|---| | Plural regimes | Firms compete by governance wrapper as much as capability | Divergent contracts, national variants, model-specific exports[^46][^35] | | Coercive absorption | State procurement dominates constraint authorship | GSAR-style clauses become standard; refusal narrows[^1][^7] | | Hybrid modularization | States bargain for specific models/enclaves, not total firm compliance | Mythos-style carve-outs and cloud-mediated segmentation[^30][^29] | | Hostile bifurcation | Distinct civilizational AI stacks harden with incompatible constraint logics | Export controls, sovereign clouds, governance arbitrage intensify[^35][^42] | ## What Would Falsify This Thesis [F] The substrate-migration thesis would be weakened if the Anthropic dispute ultimately proves to have been a narrow delivery, security, or performance dispute with no durable relationship to model-governance constraints. [F] It would also weaken if the courts ultimately treat the matter as routine procurement discipline without recognizing any constitutional-adjacent issues around retaliation, speech, or embedded refusal rights. [F] Another falsifier would be evidence that major frontier vendors all hold materially identical constraints in both public posture and contractual practice, making Anthropic’s distinctiveness largely theatrical.[^17][^2][^4][^36][^8][^9][^10] [F] The broader thesis would also be overstated if contracts, model cards, cloud terms, and safety clauses prove operationally irrelevant—meaning sovereign users can routinely disable, bypass, or ignore them without bargaining, litigation, or substitute procurement. [F] Finally, it would weaken if the current moment resolves into ordinary statutory governance quickly enough that procurement and infrastructure cease to function as de facto constitutional prototypes.[^5][^6][^42][^8][^38][^1] ### Anthropic seam falsifiers - [F] Evidence that the designation rested chiefly on ordinary security, fraud, delivery, or clearance concerns rather than governance friction.[^10] - [F] Evidence that Anthropic’s red lines were not operationally binding in products or contracts.[^2][^11] - [F] Evidence that Mythos discussions show practical convergence rather than differentiated admissibility.[^29][^30] - [F] Evidence that competitors held the same constraints in the same contractual form and faced no meaningful pressure because the issue was unrelated to constraint authorship.[^8][^9] - [F] Evidence that the injunction is later reversed on grounds that eliminate the retaliation and pretext analysis entirely.[^4][^27] ## Open Questions Worth Further Prospecting [OQ] What exact contractual or enclave-level terms governed Claude Gov deployments before the rupture, and which clauses became unacceptable to federal counterparts? [OQ] Did the government’s real objection concern autonomous lethality, domestic surveillance, or operational delay caused by meaningful human-review requirements in time-sensitive contexts? [OQ] How much of future AI governance will be negotiated at the firm level versus the model-instance, cloud-enclave, or workflow-permission layer?[^50][^12][^11][^30][^14][^42][^2][^29] [OQ] Will GSA’s draft clause survive intact, and if so, will it become the clearest federal attempt yet to seize authorship over machine refusals in procurement space? [OQ] Are courts prepared to develop doctrine for entities that are simultaneously vendors, speakers, critical infrastructure nodes, and judgment systems? [OQ] If licensed plurality emerges, what institution will decide which governance wrappers are admissible for alliances, intelligence sharing, and cross-border cloud deployment?[^37][^46][^35][^26][^7][^36][^9][^1][^4] --- ## References 1. [GSA's New AI Clause: Major Changes for AI Procurement - Our Take](https://ourtake.bakerbotts.com/post/102mnj1/gsas-new-ai-clause-major-changes-for-ai-procurement) - The clause overrides commercial terms of service, claims government ownership of custom developments... 2. [Statement from Dario Amodei on our discussions with ... - Anthropic](https://www.anthropic.com/news/statement-department-of-war) - We were the first frontier AI company to deploy our models in the US government's classified network... 3. [Military AI Policy by Contract: The Limits of Procurement ... - Lawfare](https://www.lawfaremedia.org/article/military-ai-policy-by-contract--the-limits-of-procurement-as-governance) - Key contractual terms governing military AI surveillance are now being publicly rewritten through a ... 4. [Judge grants Anthropic preliminary injunction but Pentagon CTO ...](https://breakingdefense.com/2026/03/judge-grants-anthropic-preliminary-injunction-but-pentagon-cto-says-ban-still-stands/) - "The record strongly suggests that the reasons given for designating Anthropic a supply chain risk w... 5. [AI Risk Management Framework | NIST](https://www.nist.gov/itl/ai-risk-management-framework) - NIST has developed a framework to better manage risks to individuals, organizations, and society ass... 6. [Mapping the AI Governance Landscape: April 2026 Update - CSET](https://cset.georgetown.edu/article/mapping-the-ai-governance-landscape-april-2026-update/) - MIT AI Risk Initiative researchers expanded their pipeline with CSET to map over 1,000 AI governance... 7. [The GSA's Draft AI Clause Is Governance by Sledgehammer | Lawfare](https://www.lawfaremedia.org/article/the-gsa-s-draft-ai-clause-is-governance-by-sledgehammer) - The General Services Administration's draft AI clause gets the governance problem right—then blows r... 8. [Our agreement with the Department of War | OpenAI](https://openai.com/index/our-agreement-with-the-department-of-war/) - This language makes explicit that our tools will not be used to conduct domestic surveillance of U.S... 9. [Google, Pentagon discuss classified AI deal, the Information reports](https://www.reuters.com/technology/google-pentagon-discuss-classified-ai-deal-information-reports-2026-04-16/) - Alphabet's Google is negotiating an agreement ​with the Department of Defense that would ‌allow the ... 10. [Anthropic Supply Chain Risk Designation Takes Effect - Mayer Brown](https://www.mayerbrown.com/en/insights/publications/2026/03/anthropic-supply-chain-risk-designation-takes-effect--latest-developments-and-next-steps-for-government-contractors) - The DoW has invoked two legal authorities to support its supply chain risk designation: (i) 10 U.S.C... 11. [Claude Gov models for U.S. national security customers - Anthropic](https://www.anthropic.com/news/claude-gov-models-for-u-s-national-security-customers) - We're introducing a custom set of Claude Gov models built exclusively for U.S. national security cus... 12. [Anthropic awarded $200M DOD agreement for AI capabilities](https://www.anthropic.com/news/anthropic-and-the-department-of-defense-to-advance-responsible-ai-in-defense-operations) - Claude Gov models, which we custom built for national security customers, already power deployments ... 13. [Anthropic - Wikipedia](https://en.wikipedia.org/wiki/Anthropic) - Anthropic was founded in 2021 by former members of OpenAI, including ... Jack Clark: a co-founder an... 14. [Anthropic boss rejects Pentagon demand to drop AI safeguards - BBC](https://www.bbc.com/news/articles/cvg3vlzzkqeo) - The act essentially gives a US president the authority to deem a given company or its product so imp... 15. [What the Defense Production Act Can and Can't Do to Anthropic](https://www.lawfaremedia.org/article/what-the-defense-production-act-can-and-can't-do-to-anthropic) - It has been extended many times since its enactment, most recently through September 2026. The DPA a... 16. [Anthropic, the Pentagon, and the Defense Production Act](https://reason.com/volokh/2026/02/25/anthropic-the-pentagon-and-the-defense-production-act/) - The government might now demand that Claude, Anthropic's frontier AI model, be provided without thos... 17. [Pentagon reviewing Anthropic partnership over terms of use dispute](https://thehill.com/policy/defense/5740369-pentagon-anthropic-relationship-review/) - The Wall Street Journal reported last week that Claude was used through Anthropic's partnership with... 18. [Anthropic ditches its core safety promise in the middle of an AI red ...](https://www.cnn.com/2026/02/25/tech/anthropic-safety-policy-change) - Anthropic, a company founded by OpenAI exiles worried about the dangers of AI, is loosening its core... 19. [ICYMI: Developments in Anthropic Challenges to Department of War ...](https://www.wileyconnect.com/icymi-developments-in-anthropic-challenges-to-department-of-war-supply-chain-risk-designation) - To date, the Government has not issued an exclusion order related to Anthropic or its products under... 20. [Anthropic officially designated a supply chain risk by Pentagon - BBC](https://www.bbc.com/news/articles/cn5g3z3xe65o) - The US has officially deemed artificial intelligence (AI) firm Anthropic a supply chain risk, settin... 21. [Hegseth declares Anthropic a supply chain risk, restricting military ...](https://www.cbsnews.com/news/hegseth-declares-anthropic-supply-chain-risk/) - Defense Secretary Pete Hegseth deemed artificial intelligence firm Anthropic a "supply chain risk to... 22. [OpenAI reaches deal to deploy AI models on U.S. ... - Reuters](https://www.reuters.com/business/openai-reaches-deal-deploy-ai-models-us-department-war-classified-network-2026-02-28/) - OpenAI reaches deal to deploy AI models on U.S. Department of War classified network. By Reuters. Fe... 23. [OpenAI details layered protections in US defense department pact](https://www.reuters.com/business/media-telecom/openai-details-layered-protections-us-defense-department-pact-2026-02-28/) - OpenAI said on Saturday that the agreement it struck a day ago with the Pentagon to deploy technolog... 24. [Pentagon Designates Anthropic a Supply Chain Risk - Mayer Brown](https://www.mayerbrown.com/en/insights/publications/2026/03/pentagon-designates-anthropic-a-supply-chain-risk-what-government-contractors-need-to-know) - On February 27, 2026, President Donald Trump directed all federal agencies to cease using Anthropic'... 25. [GSA AI Procurement Rules Would Introduce New Disclosure and ...](https://www.gibsondunn.com/gsa-ai-procurement-rules-would-introduce-new-disclosure-and-use-rights-requirements-for-federal-contractors/) - GSA's draft AI clause would mark a significant shift in federal procurement practices by imposing ne... 26. [GSA Releases Draft AI Clause Ahead of Upcoming MAS Refresh](https://www.burr.com/government-contracting/gsa-releases-draft-ai-clause-ahead-of-upcoming-mas-refresh) 27. [Government actions against Anthropic are 'classic First Amendment ...](https://reason.com/2026/03/30/government-actions-against-anthropic-are-classic-first-amendment-retaliation/) - In a March 26 order, Lin issued a preliminary injunction order that prohibits the federal government... 28. [Judge blocks DOD's ban on Anthropic, calls it First Amendment ...](https://www.washingtontechnology.com/companies/2026/03/judge-blocks-dods-ban-anthropic-calls-it-first-amendment-retaliation/412451/) - The court finds the Pentagon's supply chain risk designation was punishment for public criticism, no... 29. [Anthropic Is Speaking to Government About Mythos, Co-Founder Says](https://www.forbes.com/sites/zacharyfolk/2026/04/13/anthropic-is-talking-to-us-government-about-mythos---despite-tensions/) - Anthropic co-founder Jack Clark said the company was speaking to the U.S. government about its new A... 30. [Federal agencies skirt Trump's Anthropic ban to test its advanced AI ...](https://www.politico.com/news/2026/04/14/anthropic-mythos-federal-agency-testing-00872439) - They expressed frustration that the government isn't deploying Mythos more aggressively to secure it... 31. [AI Standards | NIST - National Institute of Standards and Technology](https://www.nist.gov/artificial-intelligence/ai-standards) - Incorporation of the AI RMF in international standards will further the Framework's value as a resou... 32. [Governance of Agentic Artificial Intelligence Systems - Mayer Brown](https://www.mayerbrown.com/en/insights/publications/2026/02/governance-of-agentic-artificial-intelligence-systems) - Agentic artificial intelligence (AI) systems present unique opportunities and challenges for organiz... 33. [[PDF] Building the Analytical Infrastructure for Governing Frontier AI ...](https://techpolicyinstitute.org/wp-content/uploads/2026/03/AI_Governance.pdf) - Its outputs would include assessments of whether safety frameworks actually constrain behavior, form... 34. [Pentagon's Anthropic Designation Won't Survive First Contact with ...](https://www.lawfaremedia.org/article/pentagon's-anthropic-designation-won't-survive-first-contact-with-legal-system) - On Feb. 27, Defense Secretary Pete Hegseth designated Anthropic—the maker of the AI model Claude—a s... 35. [The AI Governance Arbitrage | United Nations University](https://unu.edu/article/ai-governance-arbitrage) - Fragmented governance could distort competition, weaken international trust in AI systems and lead t... 36. [Contracts and the First Amendment, Anthropic v. U.S. Department of ...](https://www.contractsprofblog.com/2026/04/contracts-and-the-first-amendment-anthropic-v-u-s-department-of-war/) - U.S. Department of War, the District Court for the Northern District of California granted Anthropic... 37. [IPWatchdog LIVE: From 'Sneaky AI' to 'Ontology', What IP Attorneys ...](https://ipwatchdog.com/2026/03/23/ipwatchdog-live-sneaky-ai-ontology-what-ip-attorneys-need-know-ai-contracting/) - As artificial intelligence adoption accelerates across both commercial and government sectors, tradi... 38. [Breaking the deadlock on AI governance | 02 Barriers to global AI ...](https://www.chathamhouse.org/2026/03/breaking-deadlock-ai-governance/02-barriers-global-ai-governance) - International AI governance is at risk of failure. Rapid geopolitical change, institutional weakness... 39. [Pentagon Officially Flags Anthropic as Supply Chain Risk](https://deeperinsights.com/news/pentagon-flags-anthropic-as-supply-chain-risk/) - The Pentagon flags U.S. AI company Anthropic as a supply chain risk over ethical disputes on surveil... 40. [What Sparked the Pentagon's Unprecedented "Supply-Chain Risk ...](https://www.kavout.com/market-lens/what-sparked-the-pentagon-s-unprecedented-supply-chain-risk-label-for-anthropic) - The Pentagon labeled Anthropic a "supply-chain risk" due to disagreements on AI use. This impacts An... 41. [What is Anthropic's Mythos Model and Why is the Pentagon Dispute ...](https://drainpipe.io/knowledge-base/what-is-anthropics-mythos-model-and-why-is-the-pentagon-dispute-reshaping-government-ai-procurement/) - Anthropic's Mythos model is redefining AI capability, but a Pentagon clash over usage limits is resh... 42. [AWS GovCloud (US) - Amazon Web Services](https://aws.amazon.com/govcloud-us/) - AWS GovCloud (US) is an innovative compliant cloud solution that technology leaders trust to host se... 43. [How AI Safety Institutes Inform AI Governance | TechPolicy.Press](https://techpolicy.press/from-safety-to-innovation-how-ai-safety-institutes-inform-ai-governance) - A new report unpacks the first wave of AI Safety Institutes and what lies ahead for global AI govern... 44. [Mapping the Discourse on AI Safety & Ethics - AI • Objectives • Institute](https://ai.objectives.institute/blog/mapping-the-discourse-on-ai-safety-amp-ethics) - Using our prototype discourse visualization tool Talk to the City, we mapped Twitter conversations a... 45. [CNAS Insights | Setting the Rules for AI Warfare](https://www.cnas.org/publications/commentary/setting-the-rules-for-ai-warfare) - Militaries will need rules for how they adopt AI, not because autonomy is inherently illegal or unet... 46. [Pluralism in AI Governance: Toward Sociotechnical Alignment and ...](https://arxiv.org/abs/2602.15881) - This paper examines the challenge of embedding public values into national artificial intelligence (... 47. [Building AI for the Democratic Matrix: A Technical Research Agenda ...](https://knightcolumbia.org/content/building-ai-for-the-democratic-matrix-a-technical-research-agenda-for-normative-competence-and-normative-institutions-1) - To maintain democratic resilience, it is essential to build AI agents capable of choosing behaviors ... 48. [OpenAI Shares Language From Contract With the Department of ...](https://www.businessinsider.com/openai-shares-contract-language-department-of-war-mass-surveillance-weapons-2026-2) - OpenAI shared some contract language from its agreement with the Department of Defense. Its tech can... 49. [AI firm Anthropic rejects unrestricted US military use - DW.com](https://www.dw.com/en/ai-firm-anthropic-rejects-unrestricted-us-military-use/a-76143608) - Anthropic fears the unrestricted military use of its AI systems by the US government may harm democr... 50. [Anthropic rejects Pentagon's AI demands - POLITICO](https://www.politico.com/news/2026/02/26/anthropic-rejects-pentagons-ai-demands-00802554) - The artificial intelligence startup Anthropic on Thursday rejected the Pentagon's demand for unfette... 51. [The Compulsion Paradox and the Pentagon-Anthropic Spat](https://www.joneswalker.com/en/insights/blogs/ai-law-blog/the-compulsion-paradox-and-the-pentagon-anthropic-spat-when-deregulation-means.html) - This is the second installment in our series examining the Pentagon-Anthropic standoff and its impli... 52. [Pentagon rolls out GenAI platform to all personnel, using Google's ...](https://breakingdefense.com/2025/12/pentagon-rolls-out-genai-platform-to-all-personnel-using-googles-gemini/) - Other “frontier AI capabilities” will join Gemini on the new GenAi.mil platform, meant to make gener... 53. [The Inevitable Presidential AI Model | Lawfare](https://www.lawfaremedia.org/article/the-inevitable-presidential-ai-model) - An Oval Office-centered model could reshape decision-making, records, and constitutional limits on e... 54. [AI Is Changing How Stories Are Developed - And Who Decides ...](https://www.forbes.com/sites/cathyrubin/2026/03/09/ai-is-changing-how-stories-are-developed---and-who-decides-what-gets-made/) - Microdramas, predictive modeling and generative AI are moving audience data upstream — reshaping how... 55. [Media cultivation of public outlooks on artificial general intelligence](https://journals.sagepub.com/doi/10.1177/14614448261426383) - Recent titles such as Black Mirror, Blade Runner 2049, Ex Machina, Her, The Creator, and Westworld o... 56. [Westworld vs Ex Machina - Metavert](https://www.metavert.io/compare/westworld-vs-ex-machina) - Westworld vs Ex Machina: How HBO's AI theme park and Garland's Turing test thriller explore consciou... 57. [The Anthropic Cofounders Standing Up To Trump - Forbes](https://www.forbes.com/sites/richardnieva/2026/03/04/the-anthropic-cofounders-standing-up-to-trump/) - Anthropic's very beginnings came from a schism with OpenAI over AI safety — a disagreement that saw ... 58. [White House National AI Policy Framework Calls for Preempting ...](https://www.crowell.com/en/insights/client-alerts/white-house-national-ai-policy-framework-calls-for-preempting-state-laws-protecting-children) - The White House framework seeks to preempt “cumbersome” state AI laws in favor of a single national ...