The Lyceum: AI Intelligence Brief — Jun 05, 2026

The next 72 hours are the decisive legislative window for whether statutory constraints on DoD autonomous targeting enter the FY2027 NDAA. The Senate Armed Services Committee (SASC) marks up next week with two of its members — Sen. Gillibrand and Sen. Slotkin — confirmed to offer separate AI amendment tracks; assess moderate confidence that at least one provision reaches a recorded vote, based on named-source reporting and confirmed senatorial intent, but low confidence on survival given the Republican majority and direct tension with the June 2 EO calling for AI "deployed rapidly to confront any and all threats." That EO published in the Federal Register this morning (Doc. 2026-11415), locking in a voluntary governance architecture and starting a July 2 clock for a Treasury-led AI cybersecurity clearinghouse that defense press has not flagged. Meanwhile, House Armed Services Committee (HASC) FY27 text (marked up June 4) contains Sections 1512, 1513, and 1521 that would time-box AI accreditation, rewrite DoD Directive 3000.09 to cover decision-support AI, and force a classified-network modernization roadmap — moving the governance perimeter "from the trigger to the targeting stack." The cumulative implication: capability and procurement are outrunning governance, and the binding question for program managers is not bill text but whether SASC Chairman Wicker permits a floor vote.

[INTEGRATION][AUTONOMY] Palantir Maven Smart System program-of-record transition: MSS — fusing intelligence sources into one targeting interface, now 20,000+ users across 35 tools and three classification domains under a $1.3B ceiling through 2029 — is moving to a formal program of record (DefenseScoop). Decision implication: locks MSS into PPBE governance, reducing CDAO flexibility to introduce competitors and raising vendor-lock exposure ahead of the next contract-modification window.

[COMPUTE][INTEGRATION] DoD FY2027 AI Arsenal Initiative — $29.5B compute request: The largest single AI-infrastructure ask in DoD history funds centralized, secured supercomputing (DefenseScoop). Decision implication: this is the backstop for Pace-Setting Project demonstrations due July 2026; Senate markup disposition determines whether that timeline holds.

[FOUNDATION][INTEGRATION] Google releases Gemma 4 12B (June 3): A 12B-parameter open-weight multimodal model running text/image/audio/video on 16GB consumer hardware under Apache 2.0. Decision implication: lowers the bar for "open-like" multimodal deployment inside JWCC and on-prem clusters, raising immediate data-handling and model-assurance questions for edge AI experimentation.

[ACQUISITION][AUTONOMY] Lockheed GRIZZLY™ C-UAS live intercept at Yuma: A JAGM launched from a containerized system intercepted a Group 3 one-way attack drone. Decision implication: gives force-protection acquisition a modular, transportable C-UAS option to write into expeditionary solicitations without a fixed-launcher footprint.

[ACQUISITION][AUTONOMY] Air Force awards GE Aerospace and Rolls-Royce ACP engine contracts: Medium-thrust propulsion for Autonomous Collaborative Platforms moves from requirement to funded execution. Decision implication: retires a primary schedule risk for uncrewed combat aircraft and should trigger earlier integration and test-infrastructure planning.

[FOUNDATION][GOVERNANCE] NIST CAISI remains a live evaluation channel: Continued unclassified evaluation work (including DeepSeek V4 Pro) and GSA coordination. Decision implication: agencies retain a functioning technical-review layer, but it does not settle deployment, custody, or classified-use authority.

For program managers writing FY27 contract language, the operative variable is procedural: whether Chairman Wicker allows a recorded vote, not the bill text itself. Gillibrand's Secure and Accountable Military AI Act would require written approval from an Under Secretary of Defense or the Vice Chairman of the Joint Chiefs before deploying "high-consequence" AI (lethal targeting, nuclear C2), with 15-day congressional notification, and would force frontier contractors to report model-weight theft or data poisoning within 72 hours (section-by-section). Slotkin's AI Guardrails Act would bar autonomous lethal force without a human in the loop (The Hill). The contractor incident-reporting clause is most likely to survive markup — it maps onto existing DFARS cyber-reporting and carries bipartisan appeal. The autonomous-targeting prohibition is most likely stripped. (Contested — the June 2 EO, signed the same day Gillibrand proposed her safeguards, calls for AI "deployed rapidly to confront any and all threats" (Defense One), giving SASC Republicans a direct executive-policy argument that the amendment is operationally incompatible.) No Republican co-sponsor has been named — the threshold survivability question.

The program-of-record designation converts an OTA-speed deployment into a PPBE-governed program, raising both acquisition-rigidity and oversight stakes. Janes reporting (May 2026) indicates MSS can be automated so a human operator approves a target and the system connects sensor to effector; Palantir's UK defense lead confirmed adaptability "to the extent that it wants to automate targeting" (Janes). If accurate, MSS may already operate at the boundary of DoD Directive 3000.09's human-judgment requirement — a question the program-of-record designation does not resolve and the Gillibrand amendment would directly implicate. (Contested — former senior defense officials cited by DefenseScoop assess integration speed likely introduced unaudited security gaps; DoD has published no formal T&E report on MSS 3000.09 compliance.) The NDAA 2025 Section 1061 waiver-disclosure report (due December 31, 2025) remains unconfirmed as submitted — the missing baseline for any compliance assessment.

The HASC CITI print (marked up June 4, echoed by Chairman Don Bacon) is a quieter but structurally significant signal for platform owners and authorizers. Section 1512 would force a time-boxed "AI Model Rapid Deployment Framework" with CDAO-owned process metrics; Section 1513 would direct a 3000.09 rewrite within one year to cover AI that "supports, recommends, or materially influences" force-employment decisions — planning, target development, weaponeering; Section 1521 would require a 180-day Top Secret/SAP network modernization roadmap (HASC print). Assess moderate confidence these are authentic signals (drafted text, public markup), low confidence on final form — House text often narrows in conference. For primes and integrators, the downstream effect is having to prove where systems sit in a force-employment chain, not just safety and testing.

BIS guidance issued May 31, 2026 clarifies it continues enforcing pre-existing 2023 China advanced-chip license requirements; its AI-diffusion non-enforcement applies only to items not destined for entities with parents headquartered in Country Group D:5 or Macau. For export and compute-acquisition leads, this delineates the edges of the vacuum: licenses are still required for D:5-parented end users abroad absent an exception. The AI Diffusion replacement rule remains absent from the Federal Register at 384+ days, so allied compute-sharing, cloud-placement, and chip-routing decisions still proceed by assuming future legal tolerance rather than a settled framework — a planning hazard for international research and government-to-government compute arrangements.

[AUTONOMY][GOVERNANCE] Gap — 3000.09 waiver disclosure status. Whether the NDAA 2025 Section 1061 report was submitted by its December 31, 2025 deadline is unconfirmed. Matters because: Gillibrand/Slotkin amendments are written against the 3000.09 framework; if the report reveals undisclosed autonomous targeting authorities already in use (including within MSS), the debate shifts from prospective to retroactive compliance, raising CDAO and PM exposure. Watch for: congressional staff inquiry or IG referral; any SASC hearing question on waiver status.

[COMPUTE][INTEGRATION] Gap — PSP July 2026 demonstration readiness vs. AI Arsenal funding. Whether the seven Pace-Setting Projects have compute to execute operational (not lab) demonstrations is not publicly assessable. Matters because: if the $29.5B reconciliation funding is delayed or cut in Senate markup, the infrastructure backstop disappears and CDAO faces a service-branch credibility gap. Watch for: DepSecDef/CDAO framing of July demos as "operational" vs. "experimental"; any SAM.gov compute solicitation tied to PSP timelines.

[GOVERNANCE][ACQUISITION] Gap — June 2 EO Treasury clearinghouse design. The EO mandates a Treasury-led AI cybersecurity clearinghouse (with NCD, NSA, CISA) within 30 days — a July 2 deadline — structured as voluntary industry collaboration with no designed contracting or information-sharing architecture (FR Doc. 2026-11415). Matters because: it shapes whether DoE/NNSA labs and DoD can route vulnerability-detection grant funding (OMB tasked in the same window). Watch for: OMB grant-reprogramming notice; any Treasury implementation guidance by July 2.

• If SASC markup (week of June 8) adopts the Gillibrand contractor incident-reporting provision → every frontier vendor with a DoD classified contract (AWS, Google, Microsoft, NVIDIA, OpenAI, Oracle) faces a 72-hour model-weight-breach reporting duty; contracting officers should review DFARS gap coverage before markup concludes.
• If SASC strips both Gillibrand and Slotkin amendments without a recorded vote by June 12 → the FY27 legislative backstop for autonomous targeting is exhausted; PMs can proceed on current 3000.09 posture without near-term statutory risk.
• If BIS publishes a Federal Register pre-publication notice for the AI Diffusion replacement rule before June 30, 2026 → the 384-day vacuum ends; AUKUS and Gulf compute-sharing agreements acquire a legal baseline and Tier 2 GPU allocations require re-adjudication; watch Federal Register daily TOC and Docket BIS-2025-0001.
• If FY27 reconciliation preserves the full $29.5B AI Arsenal line through Senate floor action → Congress formally accepts AI infrastructure as warfighting infrastructure; PSP managers should treat July 2026 milestones as binding and prepare accelerated task-order issuance under CDAO Alpha-1 ($582M vehicle).

The Gillibrand/Slotkin analysis rests on high-confidence sourcing: primary bill text and section-by-section (Tier 1), Defense One and The Hill with named sources (Tier 2). The June 2 EO clearinghouse deadlines are authoritative (Federal Register primary document). Maven program-of-record claims are Tier 2 (DefenseScoop, Janes), but the 3000.09 compliance posture is inferred from capability descriptions, not official compliance documentation — moderate confidence. The $29.5B figure is Tier 2 citing Pentagon budget documents; PSP readiness is structural inference — low confidence on readiness specifically. HASC Sections 1512/1513/1521 are drawn from drafted committee text — moderate confidence as signals, low on final form pending conference.