In early February 2016, the security gate at a U.S. military base near Washington, D.C., swung open to admit a Navy doctor accompanying a pair of surprising visitors: two artificial intelligence scientists from Google.

In a cavernous, temperature-controlled warehouse at the Joint Pathology Center, they stood amid stacks holding the crown jewels of the center’s collection: tens of millions of pathology slides containing slivers of skin, tumor biopsies and slices of organs from armed service members and veterans.

Standing with their Navy sponsor behind them, the Google scientists posed for a photograph, beaming.

Mostly unknown to the public, the trove and the staff who study it have long been regarded in pathology circles as vital national resources: Scientists used a dead soldier’s specimen that was archived here to perform the first genetic sequencing of the 1918 Flu.

Google had a confidential plan to turn the collection of slides into an immense archive that — with the help of the company’s burgeoning, and potentially profitable, AI business — could help create tools to aid the diagnosis and treatment of cancer and other diseases. And it would seek first, exclusive dibs to do so.

“The chief concern,” Google’s liaison in the military warned the leaders of the repository, “is keeping this out of the press.”

More than six years later, Google is still laboring to turn this vast collection of human specimens into digital gold.

At least a dozen Defense Department staff members have raised ethical or legal concerns about Google’s quest for service members’ medical data and about the behavior of its military supporters, records reviewed by ProPublica show. Underlying their complaints are concerns about privacy, favoritism and the private use of a sensitive government resource in a time when AI in health care shows both great promise and risk. And some of them worried that Google was upending the center’s own pilot project to digitize its collection for future AI use.

Pathology experts familiar with the collection say the center’s leaders have good reason to be cautious about partnerships with AI companies. “Well designed, correctly validated and ethically implemented [health algorithms] could be game-changing things,” said Dr. Monica E. de Baca, chair of the College of American Pathologists’ Council on Informatics and Pathology Innovation. “But until we figure out how to do that well, I’m worried that — knowingly or unknowingly — there will be an awful lot of snake oil sold.”

When it wasn’t chosen to take part in JPC’s pilot project, Google pulled levers in the upper reaches of the Pentagon and in Congress. This year, after lobbying by Google, staff on the House Armed Services Committee quietly inserted language into a report accompanying the Defense Authorization Act that raises doubts about the pathology center’s modernization efforts while providing a path for the tech giant to land future AI work with the center.

Pathology experts call the JPC collection a national treasure, unique in its age, size and breadth. The archive holds more than 31 million blocks of human tissue and 55 million slides. More recent specimens are linked with detailed patient information, including pathologist annotations and case histories. And the repository holds many examples of “edge cases” — diseases so vanishingly rare that many pathologists never see them.

Google sought to gather so many identifying details about the specimens and patients that the repository’s leaders feared it would compromise patients’ anonymity. Discussions became so contentious in 2017 that the leaders of the JPC broke them off.

In an interview with ProPublica, retired Col. Clayton Simon, the former director of the JPC, said Google wanted more than the pathology center felt it could provide. “Ultimately, even through negotiations, we were unable to find a pathway that we legally could do and ethically should do,” Simon said. “And the partnership dissolved.”

But Google didn’t give up. Last year, the center’s current director, Col. Joel Moncur, in response to questions from DOD lawyers, warned that the actions of Google’s chief research partner in the military “could cause a breach of patient privacy and could lead to a scandal that adversely affects the military.”

Google has told the military that the JPC collection holds the “raw materials” for the most significant biotechnology breakthroughs of this decade — “on par with the Human Genome Project in its potential for strategic, clinical, and economic impact.”

All of that made the cache an alluring target for any company hoping to develop health care algorithms. Enormous quantities of medical data are needed to design algorithmic models that can identify patterns a pathologist might miss — and Google and other companies are in a race to gather them. Only a handful of tech companies have the scale to scan, store and analyze a collection of this magnitude on their own. Companies that have submitted plans to compete for aspects of the center’s modernization project include Amazon Web Services, Cerner Corp. and a host of small AI companies.

But no company has been as aggressive as Google, whose parent company, Alphabet, has previously drawn fire for its efforts to gather and crunch medical data. In the United Kingdom, regulators reprimanded a hospital in 2017 for providing data on more than 1.6 million patients, without their understanding, to Alphabet’s AI unit, DeepMind. In 2019, The Wall Street Journal reported that Google had a secret deal, dubbed “Project Nightingale,” with a Catholic health care system that gave it access to data on millions of patients in 21 states, also without the knowledge of patients or doctors. Google responded to the Journal story in a blog post that stated that patient data “cannot and will not be combined with any Google consumer data.”

In a statement, Ted Ladd, a Google spokesperson, attributed the ethics complaints associated with its efforts to work with the repository to an “inter-agency issue” and a “personnel dispute.”

“We had hoped to enable the JPC to digitize its data and, with its permission, develop computer models that would enable researchers and clinicians to improve diagnosis for cancers and other illnesses,” Ladd said, noting that all of Google’s health care partnerships involve “the strictest controls” over data. “Our customers own and manage their data, and we cannot — and do not — use it for any purpose other than explicitly agreed upon by the customer,” Ladd said.

In response to questions from ProPublica, the JPC said none of its de-identified data would be shared during its modernization process unless it met the ethical, regulatory, and legal approvals needed to ensure it was done in the right way.

“The highest priority of the JPC’s digital transformation is to ensure that any de-identified digital slides are used ethically and in a manner that protects patient privacy and military security,” the JPC said.

But some fear that even these safeguards might not be enough. Steven French, a DOD cloud computing engineer assigned to the project, said he was dismayed by the relentlessness of Google’s advocates in the department. Lost in all their discussions about the speed, scale and cost-saving benefits associated with working with Google seemed to be concerns for the interests of the service members whose tissue was the subject of all this maneuvering, French told ProPublica.

“It felt really bad to me,” French said. “Like a slow crush towards the inevitability of some big tech company monetizing it.”

The JPC certainly does need help from tech companies. Underfunded by Congress and long neglected by the Pentagon, it is vulnerable to offers from well-funded rescuers. In spite of its leaders’ pleas, funding for a full-scale modernization project has never materialized. The pathology center’s aging warehouses have been afflicted with water leaks and unwelcome intruders: a marauding family of raccoons.

The story of the pathology center’s long, contentious battle with Google has never been told before. ProPublica’s account is based on internal emails, presentations and memos, as well as interviews with current and former DOD officials, some of whom asked not to be identified because they were not authorized to discuss the matter or for fear of retribution.

Google’s Private Tour

In December 2015, Google began its courtship of the JPC with a bold, unsolicited proposal. The messenger was a junior naval officer, Lt. Cmdr. Niels Olson.

“I’m working with Google on a project to apply machine learning to medical imaging,” Olson wrote to the leaders of the repository. “And it seems like we are at the stage where we need to figure exactly what JPC has.”

A United States Naval Academy physics major and Tulane medical school graduate, Olson worked as a clinical and anatomical pathology resident at the Naval Medical Center in San Diego.

With digitized specimen slides holding massive amounts of data, pathology seemed ripe for the coming AI revolution in medicine, he believed. Olson’s own urgency was heightened in 2014 when his father was diagnosed with prostate cancer.

That year, Olson teamed up with scientists at Google to train software to recognize suspected cancer cells. Google supplied expertise including AI scientists and high-speed, high-resolution scanners. The endeavor had cleared all privacy and review board hurdles. They were scanning Navy patients’ pathology slides at a furious clip, but they needed a larger data set to validate their findings.

Enter the JPC’s archive. Olson learned about the center in medical school. In his email to its leaders in December 2015, Olson attached Google’s eight-page proposal.

Google offered to start the operation by training algorithms with already digitized data in the repository. And it would do this early work “with no exchange of funds.” These types of partnerships free the private parties from having to undergo a competitive bidding process.

Google promised to do the work in a manner that balanced “privacy and ethical considerations.” The government, under the proposal, would own and control the slides and data.

Olson typed a warning: “This is under a non-disclosure agreement with Google, so I need to ask you, do please handle this information appropriately. The chief concern is keeping this out