When the Pentagon wanted a giant underwater robot, it turned to a little-known factory on the Chesapeake Bay with a history of doing new things. You may have seen pictures of the results online — a sleek, gray sub called Manta Ray.

This $50 million unmanned underwater vehicle was developed by Northrop Grumman over four years and constructed at its plant outside Annapolis. That’s a source of pride for the people who work there.

But it also brings questions about military use of artificial intelligence home to Maryland, one of the nation’s top states for defense spending. It’s not just a question of can we do this — we clearly can — but should we turn over life-and-death decisions to smart machines?

“Just because we are in this technology environment in America and the Western world, it’s easy to do the hand wave and say all this is gonna be awesome,” said former Navy seal Coleman Ruiz, a senior fellow at the Naval Academy’s Stockdale Center for Ethical Leadership.

“Maybe. Maybe we’ll see.”

Manta Ray is not a secret. Pictures and video circulated by Northrop, Maryland’s largest private employer, and the Defense Advanced Research Projects Agency, or DARPA, show its wings stretching out from an almond-shaped body, a look instantly relatable to its namesake, a kite-shaped fish.

Ideas being tested include the potential to carry sensors, weapons or smaller drones and the capacity to survive long missions without human guidance — an AI undersea warship. A second company is testing a hydrokinetic system, which would harvest energy from the flow of water around the sub.

DARPA says the Manta Ray would provide Navy commanders with weapons that reduce the risks to more expensive, manned ships in maritime combat zones.

Or, in the context of what Adm. Sam Paparo, head of the US Indo-Pacific Command, told Washington Post columnist Josh Rogin last month, this is one of a breed of killer robots aimed at the Chinese navy.

“I want to turn the Taiwan Strait into an unmanned hellscape using a number of classified capabilities, I want to make the Chinese military utterly miserable for a month — which buys me the time for the rest of everything,” he said.

DARPA and Northrop teased the first photo of the completed prototype in April at its Sea Air Space conference in Prince George’s County, and then began sharing it on company social media and with defense industry journalists.

“I’m proud of our team in Annapolis — every day you show your dedication to providing innovative undersea solutions for our military,” Todd Leavitt, the Northrop vice president who heads up the Annapolis plant, wrote on LinkedIn.

Why would you tout an emerging, classified technology? One intended target for the message is China and Russia, who are developing their own technologies along similar lines. Another almost certainly is Washington.

The Navy is pouring billions into unmanned ships. One extra-large robot sub called Orca being developed by Boeing is so big that it could plant seafloor mines capable of firing torpedoes at passing submarines.

Some of those projects are over budget and behind schedule. Congress has been pushing the Navy to move faster, and the highly public nature of Manta Ray’s progress may be a response.

Yet public discussion of whether we should set killer robots loose in the sea appears to be over. The Pentagon seems to have answered to its satisfaction: Yes.

Ruiz, the former SEAL, commented about AI as part of the Stockdale Center’s exploration of how midshipmen should use AI, specifically machine learning and large language models like ChatGPT. The video series focused on using AI in academics and sports as examples.

But it also explored what midshipmen will encounter once they graduate. AI systems won’t absolve them from making hard decisions in combat or prevent them from making mistakes in the fog of war.

“How do you make morally justified decisions in conditions of uncertainty?” Ruiz said. “Do it the same way that you make morally justified decisions in conditions of certainty.”

Northrop Grumman and DARPA don’t say where the Navy might use future Manta Rays.

Yet, as Paparo put it, the “rest of everything” is U.S. military intervention in a Chinese invasion of Taiwan in 2030. The reasoning is partly about defending democracies, but it’s also economics. Taiwan supplies America with 44% of its semiconductor logic chips, the brains of electronic devices.

DARPA started this project in 2019, offering millions to companies interested in exploring the potential. Northrop was in the initial round and then won a $41.2 million contract in 2022 for the prototype.

As it does with many of its projects, the defense and aerospace giant brought in partners and spread the work on Manta Ray over several sites. The hub was the Center of Excellence for Undersea Warfare, located in a white colonnaded building next to the Bay Bridge.

Opened in 1967, the center started as a unit of Westinghouse Electric. Northrop Grumman bought it along with Westinghouse’s other defense businesses 30 years later, and today calls it the “gold standard” for this kind of work.

Engineers and physicists there have worked on everything from underwater “flying saucers” ocean explorers took 4,000 feet down to monitors checking the health of oyster beds in the bay and sensors that capture sharp images of the ocean floor. The workforce includes a lot of veterans, and defense work is the big deal.

Like Orca, Manta Ray is classified as an extra-large unmanned vehicle. Its design uses changes in buoyancy and two propellers to glide through the water. It can sit on the bottom virtually undetectable and is made from new materials intended to protect it from corrosion.

“Our team had to be creative and innovative to find solutions that would work for a glider UUV as big as Manta Ray,” structural engineer Hayley Sypniewski said in a short summary posted by the company. “We added more buoyancy engines, a bigger payload bay, an enhanced towing system and an extremely efficient and large hull system.”

Northrop broke down the completed prototype in January and shipped it to California for sea trials in February and March. Too big to launch from a ship, it was reassembled on a Navy pier and demonstrated its ability to glide, ascend and descend, turn, hover and anchor.

DARPA will work with the Navy to develop the design toward a possible production model.

Manta Ray doesn’t feature in the Stockdale Center’s video series on AI. But the final episode released last month does make clear the potential for surprise from AI technology.

Midshipman Emily Boutin, commissioned in May as a Marine Corps second lieutenant, talked about suffering a stress fracture and dropping out of her final track meet. The series director asked ChatGPT if she should go to support her friends.

It listed factors Boutin had already considered, and the AI added something no one expected.

“I’m really sorry to hear about your injury.”