New Court Filing: Pentagon Told Anthropic It Was “Nearly Aligned” — On the Same Day It Finalized the Designation
Anthropic filed two sworn declarations in California federal court on Friday, and the headline detail is striking: on March 4 — the very day the Pentagon formally finalized its supply-chain risk designation against Anthropic — DoD Under Secretary Emil Michael emailed CEO Dario Amodei to say the two sides were “very close” on the two issues the government now cites as national security threats: Anthropic’s positions on autonomous weapons and mass surveillance of Americans. The same Michael who said the sides were nearly aligned later posted on X that “there is no active Department of War negotiation with Anthropic,” and told CNBC there was “no chance” of renewed talks.
The declarations were submitted by Sarah Heck, Anthropic’s Head of Policy, and Thiyagu Ramasamy, Head of Public Sector. They argue the government’s case relies on technical misunderstandings and claims that were never actually raised during months of negotiations. The filing lands three days before a hearing before Judge Rita Lin in San Francisco. Anthropic needs the judge to believe the designations cause immediate irreparable harm — and the email is direct evidence that DoD didn’t actually believe Anthropic was an unacceptable risk on the day it said so.
Claude Code Channels: Control Running Sessions via Telegram or Discord
Anthropic shipped Claude Code Channels as a research preview on March 20. It lets developers hook a running Claude Code session to their Telegram or Discord account, so they can send prompts, check progress, and receive updates from their phone while away from the terminal. The channel plugin runs locally on your machine and polls the Bot API — no inbound port is opened, and no webhook is exposed to the public internet. Your session runs with full filesystem, MCP, and git access, and replies are routed back through the same messaging app.
Setup requires Claude Code v2.1.80 or higher and a Pro, Max, or Enterprise account. The feature is positioned as a direct answer to OpenClaw, a third-party tool that offered similar mobile control. Having it built natively into Claude Code eliminates setup friction and the security tradeoff of a third-party token relay. The plugin architecture is designed for expansion — Slack, WhatsApp, and iMessage are already community requests.
1M Context Window Now Generally Available at Standard Pricing for Opus 4.6 & Sonnet 4.6
Anthropic announced general availability of the full 1M token context window for Claude Opus 4.6 and Sonnet 4.6 as of March 13, with no beta header required and no long-context premium. Standard pricing applies across the full window: $5/$25 per million tokens for Opus 4.6 and $3/$15 for Sonnet 4.6. A 900K-token request bills at the same per-token rate as a 9K one. Previously, requests over 200K tokens incurred a 2x multiplier on input tokens.
The GA release also increases the media cap to 600 images or PDF pages per request, up from 100. The 1M window is available natively on the Claude Platform, Microsoft Azure AI Foundry, and Google Cloud Vertex AI. For teams running large-codebase analysis, full-document review, or multi-session summarization, this removes the main cost barrier that made extended context impractical at scale.
Cowork Projects: Persistent Workspaces With Files, Instructions & Scheduled Tasks
Anthropic is rolling out a Projects feature for Claude Cowork Desktop. It lets users create named project workspaces that persist between sessions — attaching a local folder, setting standing instructions, and binding scheduled tasks all to a single project context. The result is that Claude can resume work on a project without needing folder access re-established or objectives re-explained each time. Users can start a project from scratch, import an existing chat, or connect a folder on setup.
Scheduled tasks — introduced in the February 25 Claude app update — now integrate directly into project scopes. So a recurring research pipeline or daily reporting task can be scoped to a specific project’s files and instructions rather than running in a generic context. Cowork was launched in January 2026 as a research preview aimed at bringing Claude Code-style agentic workflows to knowledge workers who don’t work in a terminal. Projects is the feature that makes it feel like a real persistence layer rather than a session runner.
Claude Now Holds 24.4% of Business AI Subscriptions — and Is Gaining Fast
New data published by The Register shows Anthropic’s Claude grew business software subscriptions 4.9% month over month in February 2026, while OpenAI’s share fell 1.5%. OpenAI still leads at 34.4% to Claude’s 24.4%, but the gap has closed dramatically in less than a year. Among companies purchasing AI services for the first time, Anthropic wins roughly 70% of head-to-head comparisons against OpenAI. Ramp data shows nearly one in four businesses on its platform now pays for Anthropic — a year ago it was one in 25.
Anthropic’s February fundraising round put its annualized revenue run-rate at $14B, with independent reporting suggesting it climbed to around $19B by early March. The company closed a $30B Series G in February at a $380B post-money valuation. Daily active users have more than tripled since the start of 2026, and paid subscribers have doubled. The Pentagon dispute, counterintuitively, appears to have accelerated consumer and developer sign-ups rather than hurt them.
The “Nearly Aligned” Email Is the Most Dangerous Document in Anthropic’s Case
The March 4 email from Emil Michael is potentially devastating to the Pentagon’s legal position — not because it proves the designation was wrong, but because it shows the government didn’t believe it was necessary on the day it made it. For an injunction, Anthropic needs to show it’s likely to succeed on the merits and that it faces irreparable harm. If the DoD’s own top official thought the two sides were converging on March 4, then the “unacceptable national security risk” framing looks less like a sober security assessment and more like a political response to a negotiation breakdown.
The Pentagon’s counter-argument — that refusing to accept “any lawful purpose” contract language isn’t protected speech — is technically coherent but politically awkward. It essentially argues that a company can be blacklisted not for what it says, but for what it won’t sign. That’s a narrow line to walk before a federal judge. Monday’s hearing will tell us whether Judge Lin finds the timing of that email as suspicious as Anthropic does.