Rhombus Earned $10 Million as Military Reviews Flagged AI Shortfalls

Rhombus Power earned $10 million under a two-year U.S. Special Operations Command contract, yet military feedback found that some of its artificial-intelligence tools were not deployable or had failed to meet expectations. The contract ended in September 2025, and a command spokesman said no Rhombus tools are currently in use there.

The gap is striking because Rhombus promoted unusually ambitious forecasting capabilities. CEO Anshu Roy said the company’s “Unblinking Eye” predicted Russia’s invasion of Ukraine months in advance and, in another account, to within 12 hours. The company also said its platform predicted North Korean missile launches, Iranian attacks and confrontations along the China-India border. Those claims were not matched by comparable results in several documented military assessments reviewed by the Project On Government Oversight.

In August 2024, an Air Force Advanced Mobility Fusion Center assessment concluded that a Rhombus prediction tool’s “current state does not provide value or meet the criteria” for a minimum viable product. Analysts also struggled to trust it because “it lacks data transparency.” Pacific Command personnel separately complained in 2024 that the company had not delivered as advertised.

Forecasting claims and operational validation are different tests

A prediction that appears correct after a major event can be commercially persuasive, but military adoption requires a more demanding evaluation. Customers need to know what the system predicted, when it generated that output, how specific the forecast was, which data informed it and how often comparable predictions were wrong. Without that context, an isolated success cannot establish a dependable forecasting rate.

That is why the Air Force’s transparency finding reaches beyond presentation. Analysts must be able to judge whether an output is sufficiently reliable for its intended use, recognize its limitations and compare it with other sources. A tool that saves staff time may still fall short if users cannot inspect or appropriately interpret its results.

The National Institute of Standards and Technology’s guidance on AI trustworthiness treats validity and reliability as foundational. It also distinguishes transparency, explainability and interpretability: users need information about what happened within a system, how a result was produced and what that result means in context. NIST recommends accuracy measurements built around realistic test sets representative of expected operating conditions, paired with documented methodology.

Those principles explain why “deployable” is a higher bar than an impressive demonstration. Deployment depends on the software fitting the customer’s data, workflows, security environment and decision process while remaining monitorable over time. General Bryan Fenton, then the Special Operations Command commander, wrote after consulting his team that Rhombus tools had been effective in saving staff time, but some were “currently not deployable nor met expectations.” That feedback identifies a limited benefit without equating it to operational acceptance.

Senior access cannot substitute for a documented product baseline

Records also show how procurement access and technical evaluation became intertwined. In July 2025, as continued funding appeared uncertain, Roy asked Rhombus adviser and WestExec managing partner Michèle Flournoy to contact Fenton about fully funding the company. Flournoy emailed him with the subject line “Your AI Legacy at SOCOM.” Fenton initially expressed interest in what could be ahead with Rhombus before relaying the less favorable staff assessment.

The correspondence does not establish that access caused the contract award or overcame a formal evaluation. It does show why consistent contracting channels matter. Military officials reportedly warned Rhombus personnel on several occasions against sidestepping those channels. A disciplined process gives technical staff, contracting officers and senior leaders the same performance baseline, reducing the chance that executive enthusiasm and user findings proceed on separate tracks.

The scale increases the public-cost stakes. Beyond the $10 million earned from Special Operations Command, Rhombus obtained a separate contract with a ceiling of up to $200 million and won a multiyear foreign military sale financed by U.S. taxpayers. A contract ceiling is potential value rather than money necessarily paid, but it shows how quickly an inadequately validated capability can enter consideration for much larger commitments.

Former Rhombus vice president Douglas Woodall has separately alleged in a lawsuit that the company failed to report foreign contacts, mishandled classified information and allowed personnel to retain secure-facility access under expired Special Operations Command permissions. Rhombus said an outside law firm found no evidence corroborating those allegations, and its attorneys moved on October 8, 2026, to compel arbitration. The company and WestExec did not answer detailed questions about the products and advisory work.

For military AI buyers, the practical control is a procurement record that ties payment and expansion to measurable requirements: representative testing, traceable outputs, user acceptance and a clear deployment decision. Rhombus’s Special Operations Command engagement reached its end with $10 million paid, critical staff feedback on the record and none of its tools remaining in use at the command.

By David Whitaker — Associate editor for AMI’s aerospace and drone systems desk, translating flight systems, aircraft programs, spaceflight, and UAV developments into accessible technical stories.