Military warns EOD techs against uploading ‘highly sensitive’ bomb disposal data into generative AI systems
The military’s top EOD technology authority recently warned bomb technicians against uploading restricted technical material into generative artificial intelligence systems — including Pentagon-approved platforms and commercial offerings.
Internal correspondence reviewed by DefenseScoop echoed the warning issued by the EOD Technology Center (EODTECHCEN) in the last few weeks, but it suggested users of a secure system containing the EOD field’s most closely held information “may be attempting” to upload material from it into GenAI systems.
A spokesperson for the Naval Surface Warfare Center Indian Head Division, which supports the military’s EOD community, confirmed the overall warning, but said “there are no known incidents of portal data being used with AI platforms.”
The warning centers on the Automated Explosive Ordnance Disposal Publication System (AEODPS), a vast library used by EOD technicians from across the services to identify and “render safe” both U.S. and foreign weapons. AEDOPS contains material considered “highly sensitive,” NSWC spokesperson Josh Phillips said, to include controlled unclassified and classified material related to bomb disposal procedures.
The system is hosted on the Joint Explosive Ordnance Disposal (JEOD) portal, a centralized platform meant to help EOD techs securely access technical support, share intelligence and develop bomb disposal procedures. Phillips said top EOD technology and technical agencies posted an advisory to the Joint EOD portal over the last few weeks.
“The memo posted to the Joint EOD portal was to advise users of the potential for unauthorized data leaks to adversaries and the uncontrolled dissemination of hazardous, actionable intelligence that could be used to create or employ explosive weapons,” he told DefenseScoop.
The warning comes amid the Pentagon’s aggressive push to weave GenAI platforms into virtually every facet of military operations. Experts have warned that data guardrails around large language models, especially ones used by the military, are critical to protecting restricted information from becoming public or getting into the hands of adversaries.
AEODPS is a critical database that could have unique consequences to safety and national security if material from it fell into the wrong hands.
When new weapons are found on the battlefield — from missiles to rockets, bombs and grenades — the U.S. military analyzes them and creates technical publications for EOD techs to understand how to take them apart, a process known as render safe procedures. The resulting manuals are included in AEODPS.
“It is the bible and it is extremely closely guarded, and in many ways you can’t do the job without the information there,” Brian Castner, a former EOD officer, told DefenseScoop. “I understand the attraction of feeding it into an AI, but there’s a reason that it’s been kept separate and information is compartmentalized.”
One of those reasons includes preventing anyone other than EOD techs from knowing how (or attempting) to handle bombs. Another has to do with protecting information from falling into the hands of adversaries, Castner said.
“The reason a lot of it is EOD specific is that there’s intelligence value for other states, other countries, to know what we know about their weapons,” he said. “That could be: what weapons do we know how to take apart? What weapons do we not know how to take apart? … That’s not information you would want with anyone else.”
Phillips, the NSWC spokesperson, said the EODTECHCEN secures technical data through “strict operating protocols,” physically secure facilities, “hardened – often air-gapped – digital networks,” and personnel clearances. All qualified EOD techs from the Army, Navy, Marine Corps and Air Force are authorized to access the JEOD portal.
“Every EOD technician has a critical duty to protect classified technical data,” he said. “This requires unwavering discipline in using only approved government systems and maintaining positive control over all classified materials.”
In the immediate aftermath of GenAI.mil’s launch in December, DefenseScoop reported that troops and officials met the Defense Department’s premier GenAI hub with mixed reactions, with some concerned about the lack of guidance or training on safety, data security and responsible use. Phillips seemed to cite a 2024 responsible AI use policy as having guided the EODTECHEN’s restriction of uploading AEODPS into large language models.
Over the last year, the Trump administration’s Pentagon has prioritized dismantling barriers it says get in the way of technology acceleration.
“We will take a wartime approach to people and policies that block this progress,” Defense Secretary Pete Hegseth said last month. “Barriers to data sharing, authority to operate … test and evaluation and contracting are now treated as operational risks, not simply bureaucratic inconveniences; we are blowing up these barriers.”
Meanwhile, Phillips said the Joint EOD community is actively looking for ways to use generative AI systems to synthesize EOD technical data in an attempt to better inform planning, decision-making and give techs an “operational advantage.”
He added that the EODTECHCEN-supported AI integration is being approached in a way that “carefully manages generative AI’s inherent risks to unlock its recognized benefits, ensuring the EOD community maintains a decisive edge while upholding a secure mission and information environment.”
