How a Contractor’s Claude Agent Got Into the Pentagon

A defense contractor connected Anthropic's Claude Code to a secure Department of Defense system, WIRED reported this week, in what the newsroom's Uncanny Valley podcast describes as one of the most outlandish government uses of AI it has covered. The episode, hosted by executive editor Brian Barrett with reporters Leah Feiger and Zoë Schiffer, treats the incident as a case study in how individual employees are moving frontier models into sensitive environments faster than policy can catch up.

The detail that makes the story stand out is the arrangement itself: an individual contractor plugged what amounts to a personal AI agent into a classified-adjacent system rather than waiting for an approved, centrally governed deployment. The podcast frames it as a scoop drawn from WIRED's reporting, and uses it to open a broader discussion about guardrails around agentic tools in government settings.

That story shares the episode with Anthropic's own corporate coming-out. The company filed publicly for its IPO on Monday, submitting an S-1 that runs hundreds of pages and offers the first substantive look at its finances. Schiffer noted the filing contains staggering figures in both directions — billions of dollars in revenue, even larger billions in losses, and hundreds of billions already committed to compute. Anthropic is expected to raise more capital in the offering than any company in history, and would be the first major US company whose sole business is building AI software to reach the public markets.

The hosts walked the filing through a green-, yellow-, and red-flags framework, and the scale numbers themselves were treated as both the draw and the risk. Anthropic's valuation moved from $183 billion a little over a year ago to $965 billion in its most recent round, with some expecting a market capitalization near $2 trillion once trading begins. Feiger framed the sheer volume of money as an economic green flag in itself, recalling earlier WIRED reporting about people in the Bay Area selling homes with the promise of Anthropic stock post-IPO as a signal of how diffuse the pre-IPO froth has become.

Beyond markets, the episode turns to the midterms, arriving within days. Barrett reported that large volunteer groups across the country are organizing to protect communities from the possibility of ICE officers appearing at polling sites, which he cast as a reflection of a deeply uncertain election cycle.

What remains unsettled is how regulators, the Pentagon, and Anthropic itself will respond to the boundary-testing the scoop illustrates — whether agent deployments inside government get an explicit rulebook before the next incident, and whether the S-1's warnings about AI's existential risk land with investors pricing a company that could soon be worth close to $2 trillion.