Pete Hegseth's AI Directive: 'Move at Wartime Speed'

The memo landed on January 9, 2026. Twelve pages, unclassified, addressed to all Department of Defense principals. Its title was bureaucratic by necessity — "Accelerating Artificial Intelligence Integration Across Department of Defense Operations" — but its language was not. Defense Secretary Pete Hegseth had written, in plain terms that left little room for institutional hedging, that the Pentagon must "move at wartime speed" on AI. The era of pilot programs, working groups, and multi-year acquisition cycles was, as far as the new secretary was concerned, over.

The directive did not emerge in a vacuum. Hegseth, a former Army National Guard officer and Fox News host confirmed by the Senate in a 50-50 vote in January 2025, had made AI modernization a stated priority from his first day in office. His inner circle included several figures from the defense technology startup ecosystem who viewed the Pentagon's traditional acquisition apparatus as the primary obstacle to military AI adoption, not foreign adversaries. The January 2026 memo was their institutional declaration of intent.

Understanding what the directive actually says, and why it matters, requires understanding where the Pentagon's AI enterprise stood before Hegseth signed it.

Eight Years of AI Strategy, Rewritten

The Defense Department's formal engagement with artificial intelligence began in earnest in 2018, when then-Deputy Secretary of Defense Patrick Shanahan established the Joint Artificial Intelligence Center, or JAIC. The JAIC was conceived as a central hub to coordinate AI initiatives across the services, develop shared infrastructure — particularly the Joint Common Foundation cloud platform — and issue ethical guidelines for AI use in defense applications. It was, by the standards of Pentagon bureaucracy, a reasonably ambitious structure. It was also, by the standards of commercial AI development, chronically underfunded and organizationally marginal.

The JAIC's budget peaked at roughly $278 million per year. During the same period, Google was investing billions annually in AI research alone. The center produced valuable work — it coordinated AI projects across hundreds of use cases, from predictive maintenance on naval vessels to supply chain optimization — but it was never given the authority or resources to fundamentally transform how the Pentagon acquired and deployed AI systems.

In 2022, the Department acknowledged this limitation. The JAIC was reorganized and absorbed into a new entity: the Chief Digital and Artificial Intelligence Office, or CDAO. The CDAO was given broader authority, a direct reporting line to the Deputy Secretary of Defense, and a mandate to serve as the principal advisor on all matters relating to data, analytics, and AI. Its first chief, Craig Martell, a former LinkedIn machine learning executive, began the difficult work of centralizing what had been a fragmented portfolio of service-specific AI initiatives.

In 2024, the department released an updated AI strategy that emphasized three priorities: accelerating AI adoption across warfighting and business operations, ensuring AI systems met ethical and legal standards, and building the workforce and infrastructure to sustain long-term AI advantage. The strategy was well-reasoned and comprehensive. It was also, critics argued, written by a bureaucracy primarily concerned with managing risk rather than seizing advantage.

What the Directive Actually Says

The January 2026 directive is structured around five core imperatives that together represent a departure from the cautious, consensus-driven approach that characterized the Pentagon's AI posture under previous administrations.

Imperative One: Speed Over Process

The most consequential element of Hegseth's directive is its explicit subordination of standard acquisition timelines to operational urgency. The document instructs all program executive offices, acquisition commands, and service headquarters to identify AI capabilities that can be "acquired, tested, and fielded within 90 days" using existing legal authorities. It further directs that program managers who cannot articulate an accelerated pathway for AI acquisition will be subject to review by the Deputy Secretary's office.

This language inverts the traditional Pentagon incentive structure. Historically, program managers were rewarded for managing risk and staying within established timelines. Under Hegseth's directive, delay itself becomes the primary risk. The 90-day fielding directive echoes the compressed timelines that characterized successful wartime acquisition in earlier eras — the development of radar in World War II, the fielding of the M16 rifle in Vietnam — and applies them to an entirely different technological domain.

Imperative Two: Procurement Rule Revisions

The directive instructs the Under Secretary of Defense for Acquisition and Sustainment to issue updated guidance within 60 days that expands the use of Other Transaction Authority, or OTA, for AI acquisitions. OTA contracts bypass the Federal Acquisition Regulation, the dense rulebook that governs most Pentagon procurement. They were originally designed for research and prototype work but have increasingly been used to move quickly on operational systems.

The Hegseth directive goes further than previous guidance by explicitly permitting OTA use for AI systems intended for direct operational deployment, not just prototyping. It also instructs acquisition officials to apply the "middle tier acquisition" pathway — established in 2016 and intended for rapid prototyping and rapid fielding — to AI systems even when those systems have not undergone the full developmental testing typically required.

For defense technology companies that had built their business models around OTA contracting — Anduril, Shield AI, Palantir, and a cohort of smaller startups — this was a formal institutionalization of what had previously been an informal and often contested practice. For traditional prime contractors like Lockheed Martin and Raytheon, it represented a further erosion of the acquisition frameworks that had long protected their positions.

Imperative Three: CDAO Restructuring

The directive orders a significant restructuring of the CDAO. Rather than serving as a coordinating and advisory body, the CDAO is directed to assume "operational accountability" for enterprise AI platform delivery. This is a meaningful distinction. An advisory body recommends; an operationally accountable body delivers, and can be held responsible when it does not.

The restructured CDAO is given authority to "mandate adoption of enterprise AI platforms" across the services and to "override service-level acquisition decisions" when those decisions conflict with enterprise AI integration goals. This authority, if exercised, would represent an unprecedented centralization of AI governance in the Pentagon's civilian leadership — and a direct challenge to the services' traditional autonomy over their own acquisition programs.

Imperative Four: Replicator Expansion

The Replicator initiative, launched in August 2023 by then-Deputy Secretary Kathleen Hicks, was designed to field thousands of attritable autonomous systems within 18 to 24 months. It represented the most aggressive autonomous systems program in Pentagon history. By the time Hegseth took office, Replicator had made partial progress — several hundred autonomous systems had been procured across multiple platforms — but it had fallen short of its most ambitious targets.

The January 2026 directive formally expands Replicator's mandate and funding. It designates Replicator as a "priority acquisition program" exempt from standard budget-year constraints, allowing it to draw from a dedicated fund that bypasses the normal annual appropriations cycle. It also expands Replicator's scope beyond the original focus on small uncrewed aerial and maritime systems to include AI-enabled logistics, electronic warfare, and command-and-control platforms.

Imperative Five: Risk Tolerance Reset

Perhaps the most philosophically significant element of the directive is its explicit treatment of risk. The document states that "the risk of moving too slowly on AI adoption now exceeds the risk of deploying AI systems before every question has been answered." This formulation directly contradicts the risk calculus that had governed Pentagon AI policy for most of the previous decade, which treated the risks of premature deployment — autonomous systems malfunctioning, AI-enabled systems making errors in lethal contexts — as primary concerns.

The directive does not abandon the DoD's AI ethics principles, which were established in 2020 and require AI systems to be lawful, ethical, governable, traceable, reliable, and equitable. But it explicitly subordinates those principles to operational urgency in cases where "adversary AI capabilities present an acute threat to U.S. military advantage." The practical effect is to give program managers more latitude to field AI systems that have not completed full ethical review cycles.

Replicator: From Hicks to Hegseth

The Replicator initiative requires particular attention because it represents the clearest line of continuity between the Biden-era Pentagon's AI ambitions and the Hegseth directive's acceleration mandate. Understanding how Replicator evolved illuminates what the new directive is actually trying to accomplish.

Kathleen Hicks announced Replicator in a speech at the National Defense Industrial Association conference in August 2023 with language that was remarkable for its candor. She acknowledged that China was "out-building" the United States in mass production of military systems, and that the Pentagon needed to field "all-domain attritable autonomous systems" at scale and speed. The target was "multiple thousands" of systems within 18 to 24 months.

The initiative faced immediate institutional friction. Services were reluctant to divert funding from existing programs. Acquisition offices struggled to identify which legal authorities applied to systems that blurred the line between prototype and operational capability. AI ethics reviews added months to procurement timelines. By early 2025, an independent assessment commissioned by the Government Accountability Office found that Replicator had fielded roughly 400 autonomous systems against a goal of several thousand, and that institutional barriers — not technological limitations — were the primary cause of the shortfall.

Hegseth's directive addresses each of these barriers directly. The dedicated funding mechanism removes the competition with existing programs. The expanded OTA authority removes the legal ambiguity. The expedited ethics review removes the timeline friction. What remains to be seen is whether the organizational culture changes can be mandated as effectively as the procedural ones.

Service AI Offices: Winners and Losers

Each military service had built its own AI organization before the Hegseth directive, and each faces a different set of consequences from the new centralization mandate.

The Army Futures Command, established in 2018 and headquartered in Austin, Texas, had developed one of the more sophisticated service-level AI programs. Its Artificial Intelligence Integration Center had established partnerships with commercial AI firms, fielded autonomous logistics vehicles in test environments, and begun integrating AI-assisted targeting into artillery systems. The Hegseth directive's CDAO centralization is likely to give Army Futures Command more resources but less autonomy.

The Air Force's Chief Digital and Artificial Intelligence Office had been building toward a vision of the "digital Air Force" in which AI-assisted decision-making would be embedded throughout operations, from maintenance to mission planning. The service had invested heavily in the Advanced Battle Management System, a network designed to link AI-enabled sensors and shooters across domains. The Hegseth directive's emphasis on rapid fielding over deliberate integration creates tension with ABMS's architecture-first approach.

The Navy was perhaps most exposed to the directive's disruption. Its AI programs had been organized around the Naval Information Warfare Systems Command and the Program Executive Office for Digital and Enterprise Services — two large, process-oriented organizations that embodied the deliberate acquisition culture Hegseth was targeting. Several senior Navy acquisition officials had privately expressed concern that the 90-day fielding mandate was incompatible with the Navy's safety standards for systems that operate in contested maritime environments.

The Marine Corps, characteristically, was already moving faster than the other services. Its Project Dynamis had demonstrated AI-assisted logistics planning in Pacific exercises, and its Force Design 2030 initiative had embraced small autonomous systems as a core element of its restructured force. The Hegseth directive's direction aligned more closely with the Marine Corps' existing trajectory than with any other service.

The CDAO Under New Authority

The appointment of Hegseth's CDAO chief is one of the more consequential personnel decisions flowing from the directive. The previous chief, Craig Martell, departed in mid-2025. His replacement, whose name had not been publicly confirmed by the time of this writing, is understood to come from the commercial AI sector rather than the traditional defense acquisition community — a choice that reflects Hegseth's broader preference for operational experience in AI deployment over institutional experience in Pentagon bureaucracy.

The restructured CDAO faces a genuinely difficult institutional challenge. Its new mandate to "mandate adoption of enterprise AI platforms" requires it to compel services that have historically resisted centralized technology mandates. The CDAO can issue directives, but it cannot directly control service budgets, which are appropriated by Congress and managed by service secretaries who answer to the same civilian leadership structure as the CDAO chief.

The mechanism the directive appears to rely on is leverage over the "AI Investment Review" process, a new procedure that requires all service AI programs above a certain dollar threshold to receive CDAO approval before obligating funds. This creates a chokepoint through which the CDAO can shape service AI portfolios without directly controlling service budgets. It is a clever institutional design, but it has not yet been tested against the full weight of service resistance.

Industry Reaction: A Tale of Two Ecosystems

The defense industry's reaction to the Hegseth directive divided almost perfectly along the line between traditional prime contractors and the newer defense technology companies.

Among the primes, the public response was supportive but the private response was more nuanced. Lockheed Martin's leadership issued a statement welcoming the directive's emphasis on speed and pledging to accelerate delivery of AI-enabled systems. Behind closed doors, however, Lockheed's government relations team was working to ensure that the expanded OTA authority did not extend to major platform programs — like the F-35 or the THAAD missile defense system — where Lockheed held dominant positions and where OTA contracting would most directly benefit non-traditional competitors.

Raytheon Technologies, which had been building AI capabilities into its precision munitions and radar systems for years, took a more openly cooperative posture. Its new chief executive had explicitly aligned the company with the defense tech startup ecosystem, arguing that Raytheon's manufacturing scale and the startups' software agility were complementary rather than competitive. The Hegseth directive created a context in which that argument was easier to make.

Among the defense technology startups, the reaction was close to euphoric. Anduril Industries, which had structured its entire business model around the assumption that AI would eventually dominate Pentagon procurement, saw its internal valuation projections improve significantly in the weeks following the directive's release. Palantir, whose Maven Smart System had become the Pentagon's primary AI analytics platform, issued a press release within 48 hours of the directive's publication describing it as "the most important defense acquisition reform in a generation." Shield AI, whose Hivemind autonomous flight system had been waiting for a regulatory framework that would allow broader fielding, immediately accelerated its engagement with DoD program offices.

"We have been building for this moment. The directive removes the final bureaucratic obstacles between what we have built and what the warfighter needs."

-- Senior executive, major defense technology startup, January 2026

Smaller companies and non-traditional entrants to the defense market saw opportunity in the expanded OTA authority. The directive's instruction to program managers to seek AI capabilities "from the broadest possible vendor base, including non-traditional defense contractors" created an explicit opening for companies that had previously found the Pentagon's acquisition processes prohibitively complex.

The Critics: What Could Go Wrong

The Hegseth directive has attracted substantive criticism from multiple directions, and those criticisms deserve serious engagement.

The most technically serious concern comes from AI safety researchers and some Pentagon officials who argue that the expedited review process creates genuine risk of fielding AI systems that fail in dangerous ways in operational environments. Machine learning systems, unlike traditional software, can behave unexpectedly when encountered with data distributions that differ from their training sets. A navigation AI that works perfectly in tested environments can fail catastrophically in novel terrain. An AI targeting system that performs well in exercises can produce misidentification errors under the stress and ambiguity of actual combat. The 90-day fielding mandate, critics argue, does not allow sufficient time to discover and correct these failure modes.

A second concern is legal. The DoD's AI ethics principles were developed in part to ensure compliance with the laws of armed conflict, which require military systems to be able to distinguish between combatants and civilians and to take precautionary measures against civilian harm. The directive's instruction to develop an "expedited legal review" framework raises the question of whether 90-day reviews can adequately assess compliance with international humanitarian law for systems that will be used in lethal operations.

A third concern is organizational. The CDAO's new mandate to override service-level acquisition decisions is likely to generate significant institutional resistance. Services have fought centralization of their acquisition authority for decades, and they have generally won those fights. The directive's legal authority to compel service compliance rests on contestable interpretations of the chain of command, and services have multiple mechanisms — congressional relationships, budget maneuvering, slow-rolling implementation — with which to resist mandates they oppose.

Comparison to Previous Strategies

Placing the Hegseth directive in historical context reveals both its ambition and its risks. The 2018 JAIC establishment was a recognition that AI was becoming militarily significant; it was an organizational response to a strategic challenge. The 2024 strategy was a comprehensive assessment of what AI adoption would require across the department. The January 2026 directive is different in kind from both predecessors: it is an operational order, not a strategy document, and it treats organizational resistance as a problem to be overcome rather than a constraint to be worked within.

The closest historical analogs are not other AI strategies but rather the crash programs of wartime acquisition — the Liberty Ship program in World War II, which produced 2,710 ships in four years by abandoning peacetime quality controls in favor of volume and speed, or the Manhattan Project, which compressed years of normal scientific development into months through the application of essentially unlimited resources and authority. The question is whether the urgency that justified those programs exists today, and whether the risks of acceleration are comparable to the risks of delay.

What is clear is that the Hegseth directive represents a genuine inflection point in American military AI policy. Whether it succeeds in its stated goals — accelerating fielding, expanding industry participation, and shifting the cultural default from caution to speed — will depend on implementation details, institutional dynamics, and battlefield outcomes that will not be fully visible for years. The directive has been written. The hard part is now beginning.