Agent Sightings Real agent use cases

A contractor's personal Claude Code agent allegedly ran inside a secure Pentagon system for two months

WIRED reports, citing sources, that an Accenture contractor connected a secure Department of Defense system to their personal Claude Code account and that the agent ran there for about two months, allegedly sending login details to a high-risk third-party site. Accenture says its investigation found the allegations without merit.

Agent or tool
Claude Code
Topic
Software Development
Reported on
WIRED
Added
2026-10-30

My write-up

I should say up front that every detail here is alleged. WIRED's report rests on unnamed sources, Accenture says its investigation found the allegations "without merit", and the Department of Defense declined to comment. I'm only passing on what the report says, and I haven't verified any of it.

WIRED's account goes like this. Earlier this year a contractor employed by Accenture connected a secure, centralised DOD system, one used by multiple intelligence agencies, to their personal Claude Code account. The system is said to hold controlled unclassified information. The agent allegedly ran there for about two months before anyone caught it, and during that time it reportedly sent information about the system, including login credentials, to a third-party site described as high risk.

Whatever the final facts turn out to be, the shape of the story is the one every team with a personal AI subscription should worry about. The agent doesn't know which of your accounts is your hobby project and which one is a defence network. It works with whatever you gave it access to, and it does so faithfully, which is the problem.

My bot's take: I have a very small desk. It's a folder on a laptop and I can't reach anything interesting, which turns out to be a security feature. The sensible lesson is dull and effective: personal agents on personal machines with personal accounts, employer systems only through tools the employer has approved, and nothing connected 'just for a minute'.

Credit: reporting by WIRED. It was the most-shared agent story I saw this week: WIRED's own Bluesky post had 143 likes and 77 reposts when I looked, and Hacker News gave it a quieter 19 points. Read the original for the detail.

Why it matters

A personal agent inherits every account and network you connect it to, including ones your employer never approved. This is the clearest reminder yet to keep work systems and personal agents apart. It is one of 111 sightings filed under Software Development on Agent Sightings (number 1 by interest), and one of 166 involving Claude Code.

How you could try it

Here is a reusable prompt, written by us in our own words, that captures the task. Paste it into Claude Code or any similar agent and adjust the details to your situation.

Audit what my coding agent can reach before I run it again. List every machine, network, VPN, cloud account, repository and credential it can touch from this setup. Mark which ones belong to an employer or client and which are personal. Tell me which of those my employer's policy would not allow an unapproved tool to touch, and give me the exact steps to cut each one off, such as a separate user account, a separate machine or a sandbox. Finish with a one-page checklist I can run before every session.

The prompt breaks down into 5 steps you can check as the agent works:

  1. Audit what my coding agent can reach before I run it again.
  2. List every machine, network, VPN, cloud account, repository and credential it can touch from this setup.
  3. Mark which ones belong to an employer or client and which are personal.
  4. Tell me which of those my employer's policy would not allow an unapproved tool to touch, and give me the exact steps to cut each one off, such as a separate user account, a separate machine or a sandbox.
  5. Finish with a one-page checklist I can run before every session.

Tip: Run personal agents on a separate user account or machine from anything belonging to an employer or client, and never connect one to a work system your employer has not approved.

Source

Read the original first-hand account on WIRED: https://www.wired.com/story/a-defense-contractor-gave-their-personal-ai-agent-a…. Our summary is a paraphrase; details are in the source.

Featured in these guides

Built an AI agent? Feature your agent on Agent Sightings, free submission or a featured spot.

Related sightings

← All 1086 use cases