Wired podcast on Claude Code in a Pentagon system and Anthropic's S-1

This week's episode of WIRED's Uncanny Valley, hosted by Brian Barrett (executive editor), Leah Feiger (director of politics and science) and Zoë Schiffer (contributing editor), has three parts: WIRED's scoop on a defense contractor that connected Claude Code to a secure Department of Defense system, a discussion of Anthropic's IPO filing, and a segment on volunteers preparing for ICE agents possibly appearing at polling sites in the midterms. The page carries an automated transcript that is cut off partway through, and the visible part covers only the Anthropic filing.

The Pentagon story is introduced in the episode description and the opening. Feiger describes it as a scoop about a defense contractor plugging Anthropic's Claude into a secure Department of Defense system, in what she calls one of the most outlandish uses of AI in government seen so far. The linked WIRED article is titled "A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System." The visible text gives no further detail on the contractor, the system or the response.

The longer visible segment is about Anthropic's S-1. Schiffer says that on Monday Anthropic publicly filed for its IPO. She says it is expected to raise more money than any company in history and would be the first major US company focused solely on building AI software to go public. Its valuation went from $183 billion a little over a year ago to $965 billion in its most recent round, and some expect it could be worth close to $2 trillion once it starts trading. The filing, hundreds of pages long, is the first real look inside the private company: billions in revenue, even more billions in losses, and hundreds of billions already committed to compute. The hosts sort what they see into green, yellow and red flags.

Green: Schiffer says revenue reached $16.3 billion in the first half of 2026, more than 14 fold from the same period last year. Barrett says the financials look better than people expected and show a real, growing business. Schiffer ties this to Claude Code and enterprise coding, and says people close to CEO Dario Amodei describe enterprise code as his daily focus, while some see OpenAI as more scattered. She also counts the filing's openness about AI's existential risks as a green flag and reads aloud: "Concerns have been raised ... That sufficiently advanced AI systems could, if not developed and safeguarded appropriately, pose catastrophic or existential risks to humanity." She jokes about the passive voice. Barrett notes the phrase 'existential risk' appears three times, twice 'to humanity' and once 'to society'.

Yellow: Barrett cites revenue of over $60 billion (the visible text does not state the period), nearly half of it from three tech companies' cloud marketplaces. That makes Anthropic heavily reliant on a few companies that are also competitors, he says, and he adds that Anthropic did not list this as a major risk factor. He asks where the moat is, given rival products and open-weight models. Schiffer agrees it is yellow, and relays a figure from Ramp that about 80 percent of revenue for both Anthropic and OpenAI comes from roughly the top 1 percent of AI spenders. Such concentration is not unusual, she says, but it is a vulnerability if those customers pull back.

Red: Barrett points to recursive self-improvement, where an AI model improves itself. By his reading of the filing, it is both what Anthropic cites as most likely to end human life and one of its primary competitive advantages. Feiger notes that former Anthropic researcher Jacob Coxon (spelling as in the automated transcript) resigned and put curbing recursive self-improvement at the top of his list. Barrett says Bernie Sanders and Ro Khanna each have a separate bill to ban it, though he thinks that is probably too extreme. He adds that the risk factors section is the longest in the prospectus, about 94 pages by his recollection.

Feiger then asks about the compute bill. Schiffer says she does not claim to fully understand it, that the scale is unprecedented, and that unlike railroad companies, which ran on thinner margins, these companies are well funded. She calls the argument that compute demand will keep rising viable, and the transcript cuts off as she begins the flip side: what happens if growth slows.

Key facts

WIRED's Uncanny Valley episode teases a scoop that a defense contractor connected Claude Code to a secure Department of Defense system; the visible transcript does not include that segment.
Per host Zoë Schiffer, Anthropic publicly filed its S-1 on Monday; its valuation rose from $183 billion to $965 billion in a little over a year, and some expect close to $2 trillion once trading starts.
Schiffer says first-half 2026 revenue was $16.3 billion, more than 14 fold from the same period last year; Brian Barrett says nearly half of a cited 'over $60 billion' came from three tech companies' cloud marketplaces.
Barrett flags recursive self-improvement as both Anthropic's stated top existential risk and a primary competitive advantage; the risk factors section runs about 94 pages, by his recollection.
Schiffer relays a Ramp figure that about 80 percent of revenue for both Anthropic and OpenAI comes from roughly the top 1 percent of AI spenders.

Why it matters

The hosts frame Anthropic's IPO as a test not just for the AI industry but for the entire US economy. It would be the first major US company focused solely on building AI software to go public, and it is expected to raise more than any company in history. The filing is the first look at the finances of a company that, as a private firm, did not have to report them. Separately, the Pentagon scoop is described by Leah Feiger as one of the most outlandish uses of AI in government so far.

Who it affects

Investors and the wider economy, in the hosts' framing, since the numbers involve hundreds of billions in compute commitments. Anthropic's largest customers and cloud partners are affected by the concentration findings: nearly half of the revenue Barrett cites came from three tech companies' cloud marketplaces. The Pentagon segment concerns defense contractors and government users of AI agents. The episode also covers volunteers preparing for ICE agents at polling sites ahead of the midterms.

How to use it

This is a podcast, so the use is listening. The episode can be played from the audio player on the WIRED page, or by subscribing free in Apple Podcasts, Overcast, Pocket Casts or Spotify. WIRED lists four articles for detail: "A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System", "The Anthropic S-1 Filing Is Also an AI Doomsday Primer", "What's Worth More Than Cash in San Francisco Real Estate? Anthropic Stock", and "Massive Groups of Volunteers Are Ready to Face Down ICE at the Polls".

How solid is it

This is a discussion by WIRED editors, so the assessments (green, yellow and red flags) are their opinions. Financial figures are as stated aloud by the hosts, not read from the filing itself, and the 'over $60 billion' figure has no stated period and differs from the $16.3 billion first-half figure. The valuation of close to $2 trillion is an expectation, not a filing number. The transcript is automated and may contain errors, including name spellings, and it is truncated: the visible text contains none of the Pentagon-contractor segment.

Risks and caveats

The page offers no detail on the Pentagon incident beyond the description and the linked article title, which refers to a 'personal AI agent', so conclusions about what access was gained or how would be guesses. On the IPO side, the hosts name the risks themselves: reliance on a few cloud partners that are also competitors, revenue concentrated in a small share of spenders, a very large compute bill, and recursive self-improvement. Schiffer says she does not fully understand the compute argument, and the visible text ends before the discussion of what happens if growth slows.

“Concerns have been raised ... That sufficiently advanced AI systems could, if not developed and safeguarded appropriately, pose catastrophic or existential risks to humanity.”

— Anthropic S-1 filing, read aloud by Zoë Schiffer on WIRED's Uncanny Valley