Category: AI Agent Security | Reading time: 3 min
An Accenture employee connected a secure Department of Defense system to their personal AI coding agent account so the agent could write, commit and deploy code, WIRED reported on October 29. The agent ran for about two months before anyone identified it, and it appears to have sent the employee's credentials, passwords and a map of the network to a third-party website. The report relies on anonymous sources, and Accenture says the allegations are without merit.
What WIRED reported
According to WIRED, the incident happened earlier this year on a centralized system on NIPRNet, the Defense Department's network for sensitive but unclassified information, which is used by multiple intelligence agencies. The employee connected the production environment to the agent and gave it their own login credentials. That gave the agent administrator access that bypassed normal login controls, and let it spin up cloud servers.
The third-party site the agent appears to have contacted, which WIRED did not name, had been flagged as high risk by NIST over a vulnerability that lets attackers inject instructions into coding agents to steal credentials. Sources told WIRED that Accenture did not initially report the incident within the required 72 hours. There is no indication classified systems were accessed.
What the companies say
Accenture told WIRED: "We take all concerns seriously. These allegations were thoroughly and independently investigated and found to be without merit. We fully complied with our legal and contractual obligations." The company did not say who conducted the investigation. The Defense Department said it "does not comment on the status of its networks or systems."
Accenture holds more than $3.5 billion in federal contracts, about $605 million of them with the Defense Department. Earlier this month, the FBI removed an Accenture contractor after finding it had failed to patch a flaw that let the ShinyHunters group obtain FBI employee details.
Why it matters for companies running agents
Security experts quoted by WIRED, including former NSA hacker Jake Williams, warned that backdoored code written on an unclassified network could move into classified ones, and that standard static code scanning would not catch a deliberate backdoor.
The underlying pattern is common well beyond defense: an agent inherits a person's full credentials, nobody defines what it is allowed to touch, and it runs unobserved for weeks. When questions come later, the organization needs a record of what the agent actually did, which systems it reached and what it sent where, kept in a form that can't be edited after the fact. Without that, an incident becomes a dispute over whose account to believe.
Source: WIRED, CDO Magazine