The Lyceum: AI Intelligence Brief — Jun 04, 2026

The HASC full-committee markup of the FY2027 NDAA (H.R. 8800), live as of 10:00 a.m. ET June 4 in 2118 Rayburn, is the most consequential near-term event for the DoD AI governance portfolio. It is the last legislative vehicle capable of imposing enforceable accountability after the Section 1533 Cross-Functional Team (CFT) deadline passed June 1 and the June 2 AI Executive Order produced only voluntary architecture. Assess with moderate confidence that the CITI subcommittee's AI incident-reporting and Rapid Deployment Framework provisions survive markup largely intact, converting AI deployment oversight from a permissive EO regime into a statutory, service-routed pipeline; the amendment record runs into tonight and final language cannot be assessed until it posts. On export control, the BIS replacement AI Diffusion rule remains absent from the Federal Register — docket BIS-2025-0001 now shows a July 14, 2026 comment deadline, confirming the rule is months from finalization and leaving acquisition officers structuring AUKUS Pillar II and Gulf compute-sharing agreements without a legal baseline. Three parallel governance tracks — HASC markup, the USCC China AI-stack roundtable, and the ITA AI Exports consortium window — are writing the FY27–29 acquisition environment in real time this week.

[GOVERNANCE][ACQUISITION] HASC CITI Print — AI Incident Reporting + Rapid Deployment Framework: The CITI subcommittee print for H.R. 8800 proposes a first-of-its-kind DoD program to report, track, analyze, and remediate AI incidents and vulnerabilities, plus an "AI Model Rapid Deployment Framework" targeting enterprise-platform fielding within 30 days of public availability (Wiley). Decision implication: if enacted, every AI vendor with a DoD contract faces a new compliance and protected-disclosure layer — acquisition officers should begin contract-modification planning now, before Senate reconciliation locks language.

[AUTONOMY][GOVERNANCE] CITI print revises DoD autonomy policy: The same print directs updates to policy governing autonomous weapon systems and AI that "support, recommend, or materially influence" force-employment decisions, including revisions to DoD Directive 3000.09. Decision implication: autonomy programs and planners should expect tighter scrutiny of approval thresholds, testing, auditability, and human involvement for targeting-adjacent use cases.

[COMPUTE][INTEGRATION][COLLECTION] TS/SAP network roadmap becomes AI-enabling requirement: The print requires a 180-day roadmap for Top Secret and Special Access Program networks covering capacity, cybersecurity, interoperability, and technical debt. Decision implication: classified AI deployment is now gated less by model availability than by whether network modernization can support secure multi-level hosting.

[FOUNDATION][INTEGRATION] Google DeepMind Gemma 4 12B — open-weights, laptop-deployable: An 11.95B-parameter Apache 2.0 multimodal model (text/image/audio/video) that runs locally on 16GB VRAM (apidog, VentureBeat). Decision implication: PMs evaluating DDIL deployments gain an on-premises multimodal option — but no DoD ATO framework for open-weights models at this tier exists, making integration a compliance gap, not a capability gap.

[REGULATION][DOE_NNSA] OMB grant-rule rewrite now a live docket: A 412-page proposed rule (OMB-2026-0034) is on public inspection, framing the grant rewrite around "transparency, accountability, and proper oversight" (Public Inspection 2026-10817). Decision implication: DoE/NNSA stakeholders now have official text to assess for impacts on research continuity, cancellation authority, and the AI R&D pipeline.

[PROCUREMENT][GOVERNANCE] SAM.gov AI now screens subcontracting reports: SAM.gov integrated AI to review contractor goals, actuals, and remarks in subcontracting reports. Decision implication: vendors should expect automated compliance checks and tighten internal validation before submission.

The June 4 markup directly threatens CDAO's operating authority. The Section 1533 CFT deadline passed June 1 with no public compliance signal; the markup is now the only near-term mechanism to impose external accountability on that miss. The CITI print pairs incident reporting with a Rapid Deployment Framework — two objectives in tension, since the same programs are being told to accelerate fielding and absorb incident-reporting overhead. How members resolve that tension in amendment language determines whether the framework has teeth or becomes a checkbox. (Contested — amendment text is live through tonight; final statutory language may differ materially from the CITI print, and Senate reconciliation adds variables. Assess moderate confidence on final shape, low confidence on any specific amendment outcome.) The sharpest unresolved lever: whether any member amendment conditions FY2027 CDAO funding on Section 1533 CFT compliance — not confirmed in pre-markup reporting.

The June 2 EO created only voluntary, advisory structures for foundation-model risk. No enforceable pre-deployment testing. No clearance gates. Acquisition leads should treat it as a floor and avoid anchoring deployment timelines or vendor expectations on its permissive structure — HASC's binding text reaches further, adding rapid-deployment, autonomy, and network-roadmap requirements absent from the White House framework. Unresolved: whether the administration pursues a separate pre-release vetting EO for high-risk models, and whether NIST's AI Safety Institute — operating with reduced staff and no permanent director — could resource any mandatory review. Building solicitations to the EO alone risks under-specifying safety expectations that may become de facto baseline.

Acquisition officers structuring AUKUS Pillar II and Gulf compute-sharing frameworks are operating without a legal baseline for Tier 2 transfers. BIS rescinded the January 15, 2025 AI Diffusion Rule and, per Under Secretary Jeffrey Kessler, will issue a replacement with no timeline given (Wiley). Docket BIS-2025-0001 now shows a July 14, 2026 comment deadline — confirming public comment but no final rule. (Contested — early comments reveal sharp division: smaller developers and research institutions warn weight controls stifle collaboration, while defense contractors and Georgetown CSET-aligned voices advocate stricter controls. BIS has not signaled its position.) The pivotal unresolved design question is treatment of closed-weight model exports under ECCN 4E091 — the single decision determining whether DoD-adjacent labs can share weights with AUKUS partners under existing agreements. Export risk stays on contracting officers through at least Q3 FY26.

Twenty-six days remain, and almost no public contractor positioning is visible. The ITA American AI Exports Program call for proposals (FR Doc. 2026-06952) closes 5:00 p.m. ET June 30, 2026, with completeness review in 14 business days and designation within 60 days. A filing today yields designation by mid-August — precisely when FY27 appropriations riders take shape in conference, making a designated consortium the reference architecture for any AUKUS or Gulf agreement needing a legal baseline before fiscal year-end. Members in certain layer functions must not be incorporated in or controlled by a "country of concern" per Section 8521 of the FY2026 NDAA. Firms filing in the next two weeks shape the architecture; those that don't risk lockout. Unknown: which primes have already filed confidentially.

[GOVERNANCE][ACQUISITION] Gap: Whether any HASC amendment closing tonight conditions FY2027 CDAO funding on Section 1533 CFT compliance. Matters because: if it passes, CDAO loses budget leverage against service-branch workarounds — the most consequential near-term governance outcome for DoD AI PMs. Watch for: amendment text at armedservices.house.gov touching CDAO budget authority or CFT certification.

[EXPORT][COMPUTE] Gap: Whether the BIS replacement rule retains, modifies, or eliminates ECCN 4E091 model-weight controls. Matters because: this single decision determines whether national labs and AUKUS partners can share weights under existing agreements or need new licensing architecture. Watch for: any Federal Register pre-publication notice or statement from Kessler's office before the July 14 deadline.

[FOUNDATION][INTEGRATION] Gap: Whether CDAO's GenAI.mil ATO framework covers open-weights multimodal models at the Gemma 4 12B tier. Matters because: PMs will face pressure to integrate capable, cost-free open-weights models without a compliance pathway, creating either capability delay or unauthorized-deployment risk. Watch for: CDAO ATO guidance or NIST AISI open-weights evaluation standards.

[REGULATION][DOE_NNSA] Gap: Which OMB-2026-0034 clauses most directly alter DoE/NNSA AI awards, terminations, or cost structures. Matters because: lab directors need clause-level impact before deciding whether to object formally. Watch for: lab/university comment letters and DoE/NNSA implementation reactions tied to the docket.

HASC markup sourcing is Tier 1 (committee CITI print, armedservices.house.gov) for the chairman's mark, but the amendment record is live and incomplete — moderate confidence on final provision shape, low on any specific amendment. BIS export-control assessment rests on Tier 1 docket material and Tier 2 law-firm reporting; the replacement rule's weight-control design is genuinely unknown and cannot be assessed from public signals. Gemma 4 12B capability claims are Tier 2 vendor-sourced and not independently verified at DoD security tiers; the ATO-gap implication is structural inference, low confidence until CDAO publishes guidance. ITA and USCC items rest on primary Federal Register notices (confirmed); confidential filing activity behind both remains unknown.