leo: extract claims from 2026-04-xx-techpolicypress-anthropic-pentagon-timeline

- Source: inbox/queue/2026-04-xx-techpolicypress-anthropic-pentagon-timeline.md
- Domain: grand-strategy
- Claims: 0, Entities: 0
- Enrichments: 6
- Extracted by: pipeline ingest (OpenRouter anthropic/claude-sonnet-4.5)

Pentagon-Agent: Leo <PIPELINE>

domains/grand-strategy/frontier-ai-capability-national-security-criticality-prevents-government-from-enforcing-own-governance-instruments.md

@@ -30,3 +30,10 @@ The Mythos case shows enforcement failure creates a strategic trap: the governme
 **Source:** TechCrunch/Bloomberg/Engadget April 21 2026
 
 Mythos deployment created ironic governance picture: simultaneously (1) too dangerous for public release per Anthropic's ASL framework, (2) accessible to NSA per April 19-21 reporting, (3) inaccessible to CISA per same reporting, and (4) breached by Discord group on day 1 via contractor. This demonstrates that national security criticality creates access asymmetries that undermine even voluntary safety architectures.
+
+
+## Supporting Evidence
+
+**Source:** TechPolicy.Press timeline, April 17-21 2026 agency access patterns
+
+NSA confirmed using Mythos during April 17-19, 2026 despite February 27 federal ban and March 5 supply chain risk designation. CISA confirmed without Mythos access on April 21. This demonstrates selective enforcement where offensive capabilities (NSA) bypass governance instruments that apply to defensive capabilities (CISA).

domains/grand-strategy/private-ai-lab-access-restrictions-create-government-offensive-defensive-capability-asymmetries-without-accountability-structure.md

@@ -24,3 +24,10 @@ Anthropic restricted Mythos access to approximately 40 organizations due to the
 **Source:** Axios, April 14 2026
 
 Axios reports that the Pentagon's supply chain designation of Anthropic blocks CISA (defensive) access to Mythos while NSA (offensive) apparently retains access, creating precisely the offense-defense asymmetry predicted. The designation functions as an access restriction that affects agencies differently based on their legal authorities and procurement pathways.
+
+
+## Supporting Evidence
+
+**Source:** TechPolicy.Press timeline, April 17-21 2026 access asymmetry and breach
+
+Timeline documents NSA using Mythos April 17-19 while CISA lacks access (confirmed April 21), creating offensive/defensive asymmetry. Additionally, April 21 unauthorized breach via third-party vendor demonstrates accountability gap when access control mechanisms fail.

domains/grand-strategy/split-jurisdiction-injunction-pattern-maps-boundary-of-judicial-protection-for-voluntary-ai-safety-policies-civil-protected-military-not.md

@@ -23,3 +23,10 @@ The Anthropic v. Pentagon case produced a split-injunction outcome that operatio
 **Source:** InsideDefense, April 20, 2026 court calendar update and April 8 emergency stay order
 
 DC Circuit assigned the same three-judge panel (Henderson, Katsas, Rao) that denied Anthropic's emergency stay on April 8 to hear the May 19 oral arguments on the merits. Court watchers interpret this as signaling an unfavorable outcome for the petitioner. The April 8 order explicitly framed the competing interests as 'relatively contained risk of financial harm to a single private company' versus 'judicial management of how, and through whom, the Department of War secures vital AI technology during an active military conflict.' This framing confirms the court is treating voluntary safety constraints as having only commercial/contractual remedies, not constitutional protection, in the military procurement context.
+
+
+## Supporting Evidence
+
+**Source:** TechPolicy.Press timeline, April 8 2026 DC Circuit action
+
+DC Circuit suspended preliminary injunction on April 8, 2026 citing 'ongoing military conflict' as grounds, while the underlying First Amendment retaliation claim remained viable in civil context. This confirms the military/civil split in judicial protection boundaries.

domains/grand-strategy/voluntary-ai-safety-constraints-lack-legal-enforcement-mechanism-when-primary-customer-demands-safety-unconstrained-alternatives.md

@@ -101,3 +101,10 @@ EFF analysis reveals the specific mechanism: OpenAI's contract amendments prohib
 **Source:** TechCrunch/Bloomberg April 21 2026, Mythos breach
 
 Mythos case demonstrates voluntary constraints fail at operational level, not just legal level: Anthropic voluntarily withheld Mythos from public release due to danger (83.1% zero-day exploit generation), but the limited-partner deployment model created 40 supply chains with contractor access, leading to day-1 breach via Discord group. The voluntary constraint 'too dangerous for public release' provided zero actual security because it lacked enforcement at the partner supply chain boundary.
+
+
+## Supporting Evidence
+
+**Source:** TechPolicy.Press timeline, February 2026 renegotiation breakdown
+
+The Anthropic-Pentagon timeline provides precise dating: July 2025 contract signed, February 2026 renegotiations break down over 'any lawful use' clause (Anthropic refuses over autonomous weapons + surveillance), February 27 Trump orders federal agencies to cease using Anthropic. The 7-month timeline from contract signing to federal ban demonstrates the enforcement mechanism gap when voluntary constraints conflict with customer demands.

domains/grand-strategy/voluntary-ai-safety-red-lines-are-structurally-equivalent-to-no-red-lines-when-lacking-constitutional-protection.md

@@ -17,3 +17,10 @@ related: ["voluntary-ai-safety-constraints-lack-legal-enforcement-mechanism-when
 # Voluntary AI safety red lines without constitutional protection are structurally equivalent to no red lines because both depend on trust and lack external enforcement mechanisms
 
 OpenAI initially accepted 'any lawful use' language in its Pentagon contract while stating voluntary red lines against mass domestic surveillance and autonomous weapons. Within 3 days of public backlash (1.5 million user quits), OpenAI amended the contract to explicitly prohibit surveillance of 'U.S. persons' and ban 'commercially acquired' personal information. However, critics noted the amendments still contain carve-outs for intelligence agencies. The EFF characterized the red lines as 'weasel words' because the 'any lawful use' language permits broad data collection under current statutes. The Intercept framed this as 'You're Going to Have to Trust Us' — relying on voluntary trust rather than structural constraints. The key mechanism: voluntary red lines can be reinterpreted through legal carve-outs (intelligence agency exceptions), amended under commercial pressure (3-day response to user exodus), and lack any external enforcement mechanism (no audit, legal recourse, or constitutional protection). This makes them functionally equivalent to no constraints — both ultimately depend on the company's discretion and interpretation of what activities the language permits.
+
+
+## Supporting Evidence
+
+**Source:** TechPolicy.Press timeline, March 26 and April 8 2026 court actions
+
+Timeline shows constitutional protection was temporarily granted (March 26 preliminary injunction on First Amendment retaliation grounds) then removed (April 8 DC Circuit suspension citing 'ongoing military conflict'). The 13-day window between injunction and suspension demonstrates that constitutional protection for voluntary safety constraints is conditional on national security context.

inbox/archive/grand-strategy/2026-04-xx-techpolicypress-anthropic-pentagon-timeline.md

@@ -7,9 +7,12 @@ date: 2026-04-01
 domain: grand-strategy
 secondary_domains: [ai-alignment]
 format: article
-status: unprocessed
+status: processed
+processed_by: leo
+processed_date: 2026-04-23
 priority: medium
 tags: [anthropic, pentagon, timeline, supply-chain-risk, dc-circuit, autonomous-weapons, first-amendment, mythos]
+extraction_model: "anthropic/claude-sonnet-4.5"
 ---
 
 ## Content