Vendor Lock-In, Operational Continuity Risk, And Governance Boundary Between Law And Vendor Terms

Issue 72 Edition 2026-03-13 8 min read

General

Sources: 1 • Confidence: Medium • Updated: 2026-04-11 17:50

Key takeaways

Emil Michael claims that when reviewing prior administration AI contracts he found dozens of usage restrictions, including prohibitions on using AI for planning or executing operations that could lead to kinetic strikes.
Emil Michael says DoD acquisition is shifting away from complex RFP requirement checklists and cost-plus development toward simpler outcome-based requirements with multiple competing approaches and firm fixed-price contracting.
Upon taking office, Emil Michael found the DoD had 14 critical technology priority areas that had not materially changed in nearly a decade and were too vague to drive action.
Emil Michael claims China is stealing U.S. AI models and removing guardrails, creating an asymmetric disadvantage if U.S. restrictions persist.
Within about 90 days, DoD AI usage reportedly increased from roughly 80,000 personnel to about 1.2 million out of 3 million total personnel.

Emil Michael claims that when reviewing prior administration AI contracts he found dozens of usage restrictions, including prohibitions on using AI for planning or executing operations that could lead to kinetic strikes.
Emil Michael asserts that commercial AI models were embedded in highly sensitive U.S. military commands including CENTCOM, INDOPACOM, and SOUTHCOM under terms that created vendor lock-in.
Emil Michael argues the U.S. military must be able to use AI for lawful purposes without being constrained by a vendor's internal values document as AI becomes a general substrate.
Emil Michael says the DoD must engage enough AI companies so it is never single-threaded on one vendor again for critical capabilities.
Emil Michael contends that if AI model terms were enforced by shutdown logic, a vendor could theoretically cause AI tooling to stop mid-operation upon a perceived terms violation.
Emil Michael contends balancing civil liberties and national security in AI deployment should be set through Congress and regulation rather than by unilateral choices of AI company leadership.

Emil Michael says DoD acquisition is shifting away from complex RFP requirement checklists and cost-plus development toward simpler outcome-based requirements with multiple competing approaches and firm fixed-price contracting.
Emil Michael says he is pushing DoD culture toward faster yes decisions and faster no decisions so startups can decide whether to proceed or move on.
Emil Michael claims the main advantage defense primes retain over startups is scaled manufacturing and production capability rather than inventiveness.
Emil Michael expects startups to build scaled manufacturing and production capability over the next one to two years.
Emil Michael attributes slow DoD acquisition speed to post–Cold War defense-industry consolidation that followed a Pentagon-industry signal to slow growth, which he says left the U.S. behind as China began a major military buildup in the mid-2000s.

Upon taking office, Emil Michael found the DoD had 14 critical technology priority areas that had not materially changed in nearly a decade and were too vague to drive action.
Emil Michael reduced the DoD critical technology priorities from 14 to 6 and placed applied AI as the top priority.
Emil Michael moved the DoD Chief Digital and AI Office into his organization.

Emil Michael claims China is stealing U.S. AI models and removing guardrails, creating an asymmetric disadvantage if U.S. restrictions persist.
Emil Michael claims adversaries use AI partly to reduce reliance on human decision-making due to lower internal trust, while the U.S. aims to use AI to enhance human decision-making.
Emil Michael describes the frontier AI landscape as effectively four frontier companies competing over roughly a thousand highly valuable researchers who are frequently traded among firms.

Within about 90 days, DoD AI usage reportedly increased from roughly 80,000 personnel to about 1.2 million out of 3 million total personnel.
Emil Michael frames DoD AI use cases into three buckets: enterprise efficiency, intelligence analysis from large data repositories, and warfighting logistics/simulation.

What exactly changed when critical technology priorities were reduced from 14 to 6 (the six items, their definitions, ownership, and measurable objectives)?
How is 'DoD AI usage' defined in the reported jump from 80,000 to 1.2 million users, and what systems/log sources support it?
Which specific contract clauses constituted the 'dozens of restrictions' on AI use (including kinetic-strike-related prohibitions), and have they been amended or replaced?
What is the scope of commercial AI model embedding in sensitive combatant commands, and what were the lock-in drivers (data formats, APIs, hosting, integration tooling, security accreditation)?
Do any deployed systems have vendor-controlled runtime shutdown or suspension capability, and if so, what safeguards exist (escrow, offline operation, step-in rights, termination notice periods)?

Procurement could shift toward outcome-based requirements and firm fixed-price competition, benefiting vendors that can deliver measurable outcomes quickly and at scale while reducing advantage of cost-plus development incumbency.
Operational continuity concerns and vendor-enforced usage restrictions could drive diversification away from single-model dependency, favoring multi-vendor architectures, portability, and services that reduce lock-in drivers like proprietary APIs and integration tooling.
Centralizing digital and AI authority and narrowing technology priorities could concentrate spending into fewer, more actionable lanes, potentially accelerating awards and adoption if the new priorities translate into measurable objectives and ownership.

Release of the reduced priority set from 14 to 6 including definitions, accountable owners, and measurable objectives that map to budget lines or program solicitations.
Solicitations and awards showing simpler outcome-based requirements, multiple competing approaches, and firm fixed-price structures replacing complex checklist RFPs and cost-plus development.
Contract language updates that remove or standardize prior AI usage restrictions and add continuity safeguards such as step-in rights, offline operation options, and limits on vendor runtime shutdown capability.

No disclosed change in priorities beyond renaming, with no ownership or measurable objectives, and no observable linkage to acquisition actions or resource allocation.
Procurement continues to be dominated by complex RFP checklists and cost-plus development with limited competitive downselects, indicating the stated reform is not implemented.
Persistence of single-vendor embedding with proprietary lock-in drivers and no continuity protections, alongside unchanged restrictive clauses governing AI use in operational contexts.