Pentagon CDAO Quintuples Scale AI Contract to $500M Using No-Bid OTA — Eight Months After Original Award

On May 7, 2026, the Department of War’s Chief Digital and Artificial Intelligence Office (CDAO) announced a five-fold expansion of its enterprise agreement with Scale AI — from a $100 million ceiling to $500 million — just eight months after the original award. The expansion used Other Transaction Authority (OTA), the same no-bid mechanism that simultaneously allowed the contract to be modified without competitive solicitation. Any Pentagon component can now launch a Project Agreement against the vehicle without issuing a new bid.

Contract Architecture

The CDAO Production OTA agreement, originally awarded September 17, 2025, gave Scale AI a five-year, $100 million enterprise ceiling. It covered three platforms deployed on classified networks (NIPR, SIPR, and JWICS up to TS/SCI):

• Scale Data Engine — human-expert-labeled training data for computer vision models
• Scale GenAI Platform — fine-tuning and deploying generative AI models in classified environments
• Scale Donovan — AI decision-support platform for intelligence analysts and warfighters

Demand across the Department “exceeded the original ceiling” within months, per CDAO. Components from the Army, Navy, Marine Corps, defense agencies, and Office of Secretary of War offices all launched Project Agreements. The May 7 modification increased the ceiling to $500 million — with no new competitive bid required. Each new Project Agreement added by a Pentagon component adds contract value administratively, meaning $500 million is a ceiling, not a terminal figure.

Accountability Gaps

Three oversight mechanisms were bypassed or inapplicable under OTA:

1. No Privacy Impact Assessment required: OTA vehicles are not subject to the PIA requirements that apply to FAR-governed contracts. Scale AI’s data-labeling and AI platforms process sensitive military and intelligence data at classified levels; no PIA was filed as of announcement.
2. No competitive solicitation: FAR competitive bidding requirements do not apply to OTA. The original $100M award was not competitively bid in the standard sense; the five-fold expansion used the same vehicle.
3. No subaward transparency: OTA mechanisms do not require public subcontractor disclosure. The Project Agreements launched by Army, Navy, and Marine Corps components are not individually publicly solicited or announced.

A January 9, 2026, memo from Secretary of War Pete Hegseth directed CDAO to treat “barriers to data-sharing, authority to operate, test and evaluation, and contracting” as “operational risks.” He established a Barrier Removal Board with authority to waive any non-statutory requirement — framing PIA requirements and competitive-bid processes as bureaucratic obstacles to be removed.

Meta Connection and Data Confidentiality Risk

Scale AI is 49% owned by Meta Platforms (non-voting stake), following Meta’s $14.3 billion investment in June 2025. The transaction valued Scale AI at $29 billion. Scale AI’s founder Alexandr Wang departed the company to become Meta’s Chief AI Officer, leaving Jason Droege (former Uber Eats founder) as Interim CEO.

The Meta stake has structural implications for the Pentagon contract:

• Scale AI’s flagship defense product, Defense Llama, is built on Meta’s open-source Llama 3 framework — meaning Meta’s underlying model architecture is deployed in classified Pentagon environments
• Following the Meta deal, Google, OpenAI, and xAI all reduced or terminated data-labeling partnerships with Scale AI over data confidentiality concerns
• The Pentagon appears unconcerned by the same data-confidentiality risks that drove commercial AI labs away: a company 49% owned by the world’s largest social-media corporation now holds a $500M contract to process classified military data and AI operations

Architect Network

Cameron Stanley (CDAO Director, January 2026–present): Former lead of Project Maven under the Office of the Undersecretary of Defense for Intelligence and Security; then national security digital transformation lead at Amazon Web Services. Re-entered government service January 2026 with a team recruited from AWS, Databricks, Palantir, and Meta. Stanley’s CDAO oversees the OTA vehicle expansion. He described Maven Smart System usage surging 89% (classified) and 38% (unclassified) month-over-month during the May 2026 Iran strike campaign.

Emil Michael (Under Secretary of War for R&E / CTO, confirmed May 2025): Former Chief Business Officer at Uber; previously Special Assistant to Defense Secretary Robert Gates (2009–2011). Nominated by Trump on Inauguration Day. Oversaw the January 2026 “AI-first” strategy memo. CDAO was realigned to report under Michael’s R&E office in August 2025 — placing the OTA procurement vehicle under a political appointee with a prior Uber executive background.

Pete Hegseth (Secretary of War): No prior defense procurement or contracting background. January 9, 2026, “AI-first” memo explicitly frames accountability mechanisms as obstacles. The OTA expansion is the operational implementation of Hegseth’s policy direction.

Dan Tadross (Scale AI Head of Public Sector): Leads Scale AI’s defense business; testified before Senate Armed Services Committee in March 2025. Told Bloomberg that the Pentagon was already “pushing the limits” of the original $100M deal. Scale AI’s defense business is described as the company’s fastest-growing segment.

Alexandr Wang (Scale AI founder; now Meta Chief AI Officer): Though no longer Scale AI CEO, Wang accompanied Trump’s Gulf tour in May 2025, where Saudi and UAE AI deals were signed. Wang’s current role at Meta creates a direct line from a government contractor’s founding architecture to a social-media company with its own foreign-policy and commercial interests.

Contract Trajectory

Scale AI’s documented federal contract history shows rapid escalation:

Lock-In Architecture

The OTA vehicle’s “any component can add a Project Agreement” structure creates progressive switching-cost lock-in. Once Army, Navy, Marine Corps, and defense agency workflows are running on Scale Donovan’s decision-support infrastructure at classified levels, recompetition requires:

1. Disclosing the classified workflow architecture to new bidders
2. Rebuilding human-expert labeling pipelines at scale with a new vendor
3. Retraining warfighters and analysts on different interfaces

This is identical to the Palantir DHS lock-in pattern documented in 2026-05-13--dhs-zero-privacy-impact-assessments-palantir-1-billion-no-bid: enterprise AI vendors use initial OTA vehicles to embed themselves in classified operational workflows, then exploit switching costs to avoid recompetition indefinitely. Combined, the Scale AI CDAO ($500M) and Palantir DHS ($1B) no-bid contracts represent over $1.5 billion in surveillance-adjacent AI procurement awarded without competitive bidding in 2025–2026.

Defense Llama: Capability Claims vs. Independent Assessment

Scale AI’s Defense Llama — a national-security language model built on Meta’s Llama 3 — is available through Scale Donovan on TS/SCI networks. Scale AI marketed its airstrike-planning capabilities in 2024, but an independent assessment by The Intercept (November 2024) found:

• Trevor Ball (former Army EOD technician): called Defense Llama guidance “worthless” for targeting without building detail inputs
• Wes J. Bryant (retired Air Force targeting officer): said military units would find LLM-based weaponeering decisions “laughable and unusable”
• N.R. Jenzen-Jones (arms analyst): found the system response “generic to the point of uselessness,” using “clumsy and imprecise terminology”
• Jessica Dorsey (Utrecht University law professor): called the approach “amateurish” and potentially circumventing legal obligations

Scale AI claimed the marketed example didn’t represent actual capabilities but declined to provide evidence of actual performance. The $500M OTA expansion occurred despite these unresolved capability questions.