Shield AI V-BAT Severs Romanian Officer’s Fingers as Reuters Documents 50-Plus Crashes

A Romanian Navy official lost two fingers and fractured a third on May 12 when her hand was caught in the propeller of a Shield AI V-BAT drone during a training exercise on a boat off the Texas coast. Romania’s Ministry of National Defence confirmed the injury to Reuters, which reported it Thursday as part of a broader investigation into the company’s flagship reconnaissance aircraft.

This is the second time a V-BAT propeller has maimed a service member. In April 2024, a U.S. Navy sailor assisting a V-BAT landing had three fingers partially amputated. After that incident, Shield AI added new landing gear and warning stickers near the propeller, and modified the drone so it no longer needed a human to help it launch or land. We covered the V-BAT’s expanding weapons partnerships earlier this year, when the company was positioning the platform as combat-proven. The hardware that keeps cutting people is the same hardware Shield AI sells as one of the most operationally tested VTOL aircraft flying.

The Reuters account, built from interviews with 21 former employees, executives and investors plus a whistleblower complaint and a lawsuit, also documents a crash rate that the company’s public framing does not match. The Romanian officer had surgery to reattach her fingers on May 12 and May 16 at University Medical Center New Orleans, then deteriorated and was moved to Walter Reed National Military Medical Center in Maryland, where she remained as of May 25.

Reuters counted more than 50 V-BAT crashes in 18 months

More than 50 of roughly 200 upgraded V-BATs in Shield AI’s internal fleet have been destroyed in crashes during testing or training over the past 18 months, according to two people Reuters cited with knowledge of the matter. That is a quarter of the fleet, and it sits against a very different number the company gives customers.

Shield AI told Reuters its customers had experienced only 10 “operational mishaps” since early 2025, when the upgraded V-BAT entered service. The company did not comment on the internal-fleet crashes and did not elaborate on the customer incidents. It defended the aircraft in a statement, saying “operational mishaps are common” for a drone like the V-BAT and that the platform had logged 18,000 flight hours since 2019.

The documented failures go beyond unit losses. A V-BAT crash-landed on a runway during a NATO-led unmanned systems event in Portugal last September, per a witness and two videos Reuters reviewed. In February, Shield AI paused flights for several weeks to investigate a spate of crashes, one of which ignited a Texas grass fire that burned more than 40 acres before crews put it out. The previous July, two Shield AI employees flying a Cessna near a V-BAT during a detect-and-avoid test had to steer clear of a potential collision when the drone failed to spot their plane. One employee’s young son was aboard the Cessna.

Whistleblower alleges Shield AI hid failures from buyers

Jacob Miller, a former product manager, filed a whistleblower complaint with the Department of Labor’s Office of Administrative Law Judges in May alleging Shield AI obscured technical flaws to land military sales. He described a “fake it ’til you make it” Silicon Valley mindset “being applied to equipment that can cause severe immediate harm to people and war fighters.”

The specifics in the complaint are what make it serious. During one test, Shield AI told the Greek military a V-BAT was flying autonomously when an operator was actually piloting it manually, Miller alleged. He also said the company falsified or scrubbed data in internal mishap reports to build a “falsely favorable narrative” about V-BAT performance, and that the revised data helped secure contracts with Naval Air Systems Command along with Greece, Japan, Norway, Taiwan and Ukraine. Ukraine declined to comment to Reuters. The other militaries and NAVAIR did not respond.

Miller separately sued Shield AI and senior director Trey Lindsey in May, alleging he was fired for raising safety concerns. At least three employees who flagged safety issues over the past 18 months were fired or left the company. Last year Shield AI hired law firm Littler Mendelson to investigate hostile-work-environment and air-safety claims, though Reuters could not determine what the investigation found. Lindsey and Littler Mendelson did not respond to comment requests.

Responding to Miller’s allegations, a Shield AI spokesperson said the company could not comment on ongoing litigation, adding: “We believe these claims lack merit, and we intend to vigorously defend ourselves against this attack on Shield AI and its mission.”

Shield AI carries a $12.7 billion valuation and growing Pentagon ties

Shield AI was valued at $12.7 billion in a March funding round co-led by JPMorgan, making it one of the larger defense-tech bets to come out of Silicon Valley. Founded in 2015 by Ryan Tseng, who sold a phone-charging company to Qualcomm, and his brother Brandon, a former Navy SEAL, the firm built its pitch on loosening the grip that prime contractors like Lockheed Martin and RTX hold on Pentagon work. It acquired the V-BAT when it bought Martin UAV in 2021, and has spent the years since pitching autonomy as the core of future warfare.

The platform has real customers and real deployments. Romania’s Naval Forces signed a $30 million V-BAT agreement last year and told Reuters the contract remains in effect despite the injury. The U.S. Coast Guard put the V-BAT through operational testing aboard National Security Cutters in 2025, and the drone has flown reconnaissance missions in Ukraine through heavy electronic jamming. In February, Vice President JD Vance was shown the latest V-BAT during a tour of Armenia, telling a LinkedIn video, “Holy shit. Look at this thing.”

Shield AI said the May 12 injury came from “a violation of established safety procedures, not from a product defect,” without naming the violation. Romania’s defence ministry said it was investigating and that drawing conclusions about fault or preventability would be premature.

The X-BAT inherits V-BAT flight controls

Shield AI is now marketing the X-BAT, a larger jet-powered drone Reuters reports is expected to cost around $30 million and designed to fly alongside fighter jets as a loyal wingman. The company itself has put per-unit pricing closer to $27 million at the October reveal, in the same range as rival Collaborative Combat Aircraft. The Pentagon’s Defense Innovation Unit recently awarded the company a contract for the X-BAT, a Pentagon spokesperson confirmed to Reuters. An April pitch deck showed Shield AI requesting $500 million to develop four X-BAT prototypes by 2029, against a total program cost of $1.3 billion.

Here is the detail that connects the two aircraft. The X-BAT is expected to use the same flight controls as the V-BAT, according to a presentation Shield AI made to the Indian government last April that Reuters reviewed. We covered the X-BAT’s October reveal in Washington, where the company leaned on Hivemind’s combat record in Ukraine as the proof point for the new jet. That record runs on the flight-control stack now under scrutiny. Asked whether the Pentagon worried about the X-BAT’s reliance on V-BAT technology, the spokesperson said risk is “inherent to technology development and innovation” and called it “a critical learning process.”

DroneXL’s Take

The gap between 10 and 50 is the whole story. Shield AI tells customers its V-BAT logged 10 operational mishaps since early 2025. Reuters’ sources put internal-fleet losses at more than 50 in 18 months. Both numbers can be technically true if you define “customer mishap” narrowly enough to exclude everything that happens on your own test range, which is exactly the kind of definitional gap that lets a company say “strong safety record” and “operational mishaps are common” in the same statement. We have watched this framing harden over a year of coverage, from the Coast Guard’s perfect-score trials to the Ukraine jamming tests to the LIG Nex1 weapons deal. The marketing has been relentlessly upbeat. The internal data, if Miller’s complaint holds, was edited to stay that way. Shield AI is not alone in this gap between pitch and performance. We covered the Reuters investigation last November that found Anduril’s Altius drones nosediving during Air Force tests, and the pattern of venture-backed defense hardware shipping ahead of its reliability is now its own story.

I have flown enough hours to know that propellers do not care about your mission statement. A spinning ducted fan that took three fingers in 2024 and two more in 2026 is a hardware-human interface problem, and “violation of established safety procedures” is the explanation every operator dreads hearing, because it quietly moves the fault from the machine to the person standing next to it. Shield AI may be right that the operator erred. Romania’s defence ministry says it is too early to know. The company naming a procedure violation before its customer’s investigation concludes is itself worth noting.

The forward question is narrow and answerable. Miller’s whistleblower complaint sits with the Department of Labor’s Office of Administrative Law Judges, and his lawsuit against Shield AI and Trey Lindsey is filed. Discovery in either proceeding could surface the unedited mishap reports, and the specific allegation that Shield AI told the Greek military a manually piloted V-BAT was flying autonomously is the kind of claim that either checks out against flight logs or it does not. Whether the X-BAT’s shared flight controls become a Pentagon concern is also an open question Reuters put directly to the Defense Innovation Unit, and the answer it got, that risk is inherent to innovation, is not the same as a clean safety review. Watch the Labor Department docket. That is where the 10-versus-50 question gets resolved or buried.

Source: Reuters (reporting by David Jeans).

DroneXL uses automated tools to support research and source retrieval. All reporting and editorial perspectives are by Haye Kesteloo.