Security experts warn that AI-generated code deployed on U.S. government networks could behave differently across environments, hide malicious functions from conventional security scans and undermine confidence in critical tools. Although there is no indication that attackers penetrated sensitive or top-secret systems, the incident highlights risks ranging from internal attacks and distributed denial-of-service campaigns to intelligence or operational delays.
“It’s possible for your code to behave perfectly fine in NIPR and do malicious, private behavior in a SIPR network. It’s 100% possible,” Williams said. Extracting data from these networks is difficult, he added, but malicious code could still be used to attack systems from within. “I think the most likely scenario is some kind of distributed denial of service attack,” he said.
Even if the immediate issue is resolved, Triner said, subtler problems—including intelligence or operational delays—could cause damage and lead users to lose confidence in the tool.
The challenge of reviewing AI-generated code
As “vibecoding”—using an AI assistant to write code for a product described by an engineer—becomes more common, scrutinizing the resulting software is becoming harder. AI systems can generate far more code than humans typically produce, making the output time-consuming and difficult to review.
“When you’re using an agent, you don’t know the supply chain or even the inference chain of what the code was ultimately generated for,” Williams said. “Many of the codes that these LLMs were trained on have vulnerabilities of their own, and the same types of architectural and architectural issues that we see in other code can reoccur.”
Why conventional scanning may not be enough
The government has a method for identifying some flawed code within NIPRNet. Static Application Security Testing, or SAST, scans software for bugs and defects before allowing it to run. But Williams noted that “backdoors inserted by hackers are not vulnerabilities, even if we didn’t want them to be,” meaning they could evade SAST detection.
SAST also typically does not identify structural problems in the logical chain through which code operates. That limitation could make it difficult to detect malicious behavior that does not resemble a conventional software vulnerability.
No evidence of a breach of classified systems
There is no indication that the hackers were able to penetrate sensitive or top-secret systems. Still, Williams said the incident should prompt serious concern.
“Architecturally, we believe that any information about NIPR that cannot be verified through open source intelligence will be held in close custody by the Department of Defense, whether or not it touches classified networks,” he said.
Earlier concerns involving AI and defense systems
This is not the first time AI-related issues have left the Department of Defense vulnerable. Last year, ProPublica reported that Microsoft used “digital escorts” on projects with the federal government, with U.S. employees who held security clearances overseeing code written by outsourced workers in China.
Experts who spoke to ProPublica said the program posed special security risks because U.S. officials were often unable to vet the code themselves, leaving open the possibility that malicious code could be injected into government systems without their knowledge.
Source: www.wired.com
