Home / Articles / The contractor's agent on a Pentagon network / Versions / v1.1.0
The contractor's agent on a Pentagon network: what changed in v1.1.0
From v1.0.0 (2026-10-30, cc476b00a) to v1.1.0 (2026-10-30, this release), paragraph by paragraph.
5 paragraphs added, 0 removed, 4 changed in place, 62 unchanged. About 236 words added and 0 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
14 unchanged paragraphs, under In short {#in-short}, What was reported {#reported}
• The setup. To automate writing code and starting cloud servers, an Accenture employee connected the system to their personal Claude Code account and gave the agent their own login credentials. That gave the agent administrative access and let it bypass the security controls enforced through the normal login. ItWIRED's wroteown codewords are that the employee connected "a secure production environment" to their personal Claude account "so as to allow an AI agent to write and committeddeploy itcode". WIRED dates the incident to the"earlier systemthis on the employee's behalf.year".
3 unchanged paragraphs
The parties' answers, as quotedWIRED bygives thethem: outlets:an Accenture spokesperson said "These"We take all concerns seriously. These allegations were thoroughly and independently investigated and found to be without merit",merit. andWe that itfully complied with itsour legal and contractual obligations.obligations." WIRED adds that, when pressed, Accenture did not say who conducted the investigation, nor dispute "that there had been what could be characterized as an incident". A DOD spokesperson said "the Department does not comment on the status of its networks or systems. We take all concerns seriously."systems." Anthropic pointed WIRED to its usage policy, which forbids using its products to "Interfere with the operation of military bases and related infrastructure" and to "Bypass security controls such as authentication systems, endpoint protection, or monitoring tools without authorization" (I checked both phrases on Anthropic's policy page). WIRED reports no sign that anyone reached a classified system.
36 unchanged paragraphs, under Read through its shadow {#shadow}, The agent became the person {#insider}, An Agent Behaviour Policy in one paragraph {#abp}…
Corrected {#corrected}
The first version of this article, published on 30 October 2026, was written from WIRED's posts and six outlets, because WIRED's page could not be fetched from here. Later that day I read the article itself, and it showed two things the shadow had wrong, which this article repeated:
• A quote given to the wrong party. "We take all concerns seriously" is the first sentence of Accenture's statement, not the end of the DOD's. One outlet joined it to the DOD's line, and we copied it.
• Commits. The detail that the agent "committed" code on the employee's behalf came from secondary coverage; WIRED's own text, in the part I have read, says the setup was to let the agent "write and deploy code".
Both are corrected above, and the version history shows the change. It is a small, useful lesson for the shadow method: the shadow is evidence, and it can also carry a quote across a paragraph break.
1 unchanged paragraph, under What this does not claim {#limits}
• The incident is reported, not proven. It rests on unnamed sources in WIRED's article, which I couldfirst notread read,through asothers' summarisedsummaries byand others;then, in part, in screenshots of the article itself; Accenture says the allegations were investigated and found to be without merit, and the DOD would not comment. Nothing here says what Accenture, the employee or the DOD did or did not do beyond what was reported.
4 unchanged paragraphs, under Where this comes from {#sources}
• WIRED, "A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System", 29 October 2026, the article, not read:article](https://www.wired.com/story/a-defense-contractor-gave-their-personal-ai-agent-access-to-a-secure-pentagon-system/): its fetch wasis refused from here.here; read in part from screenshots the editor shared on 30 October 2026, for the corrections below. WIRED's posts on X: the contractor and Claude Code and two months on a DOD network.
4 unchanged paragraphs