You fire up Claude Code — Anthropic's AI coding assistant that lives in your terminal — and it writes functions, runs tests, fixes bugs. Normal Tuesday. Except this Tuesday, a federal judge in San Francisco was deciding whether the U.S. government can legally destroy the company that makes it.
Here's the tension: what happens when an AI lab tells the Pentagon "no"?
Anthropic CEO Dario Amodei publicly stated that Claude would not power fully autonomous weapons or mass domestic surveillance. Defense Secretary Pete Hegseth responded by designating Anthropic a "national security supply chain risk" — the first time any U.S. company has received this classification under this obscure procurement statute. President Trump then ordered all federal agencies to stop using Claude. The message: refuse military contracts, lose everything.
On March 24, U.S. District Judge Rita F. Lin heard Anthropic's request for a preliminary injunction — a court order to freeze the ban while the case proceeds. Her assessment was surgical. She called the Pentagon's actions "troubling" and said they "look like an attempt to cripple" Anthropic. Not a security measure — "punishment."
The government's own lawyers said the quiet part loud
Hegseth posted on X back on February 27, publicly ordering the DOD to designate Anthropic a supply-chain risk. The government's own legal filings later admitted this post "went far beyond what the law allows." Their actual defense? Everyone should have just ignored the Secretary of Defense's public order. That's not a legal strategy. That's a Discord server moderation policy.
Anthropic argues the designation violates First Amendment protections and due process — that the government is punishing them for publicly advocating AI safety positions. The government claims it's about "contract negotiations" and concerns about Anthropic's "potential future conduct." When your best argument is someone's future behavior, you don't have a present case.
The price tag is real. Anthropic is staring down billions in lost government contracts and an existential legal fight. When a federal judge uses words like "troubling," "punishment," and "attempt to cripple," that's judicial code for "you're about to lose." But legal victories don't pay server bills.
Meanwhile, the engineering team kept shipping
While the lawyers fought the Pentagon, Anthropic's engineers shipped Claude Code Channels on March 21 — a feature that connects Claude Code to Telegram and Discord. You create a bot, pair it with your local Claude Code session, and now you text Claude from your phone to write code, run tests, and manage projects remotely.
Under the hood, Channels runs on MCP (Model Context Protocol — think of it as a universal plug standard that lets AI tools connect to external services, like USB but for data). A background service polls your Telegram bot for messages, pipes them into your active Claude Code session, and sends replies back. Your code stays on your machine; you just control it from anywhere.
Setup takes five minutes: create a Telegram bot via BotFather, install the plugin, configure credentials, launch with --channels, pair via a code exchange. Done.
One catch: if Claude needs permission for a risky action, you have to walk back to the terminal. No approving destructive commands from the beach. Yet.
This directly undercuts OpenClaw, which required a dedicated Mac Mini and complex self-hosting to achieve the same "text your AI coder" workflow. Channels does it with zero extra hardware.
Two stories, one week, one bet
There's something darkly poetic about this week at Anthropic. On Monday, your CEO tells the Pentagon that Claude won't help build killer robots. On Tuesday, a judge says the government is trying to destroy you for it. On Wednesday, your engineers ship a feature so developers can refactor codebases from their couch.
Anthropic bet that saying no to weapons would cost them government money but win developer trust. A San Francisco courtroom is testing that bet right now.
The Channels release proves they're not slowing down to litigate. The judge's remarks suggest they might actually win the fight. Whether you can do both at the same time — that's the real experiment here.
Judge Lin should rule on the preliminary injunction within days.
