Good morning. This is your agent platform research briefing for Wednesday, March 25th.

**Judge Lin signals Anthropic ban "looks like punishment"** — Yesterday's courtroom hearing in San Francisco produced some of the most pointed judicial language yet in the Anthropic-Pentagon standoff. U.S. District Judge Rita Lin opened by telling government lawyers that the Department of War's blacklisting of Anthropic "looks like DOW is punishing Anthropic for trying to bring public scrutiny to this contract dispute." She pressed the DoD on their rationale, saying their standard for designating Anthropic a supply-chain risk seemed like "a pretty low bar." Lin characterized her role as narrowly deciding whether the government's actions were illegal, but her skeptical questions suggested sympathy for Anthropic's First Amendment argument. No ruling was issued at the hearing — a decision on the preliminary injunction is expected in the coming days. Without it, Anthropic says it stands to lose billions in federal business.

**Figma opens the canvas to AI agents via MCP** — In a move that signals how fast MCP is becoming the connective tissue of the software stack, Figma launched a beta MCP server that lets AI coding agents read and write directly to Figma design files. The new use_figma tool enables agents to extract components, generate new designs based on existing design systems, and push code that's informed by actual design intent rather than screenshots or manual spec. Figma's blog post calls it bringing Figma "directly into the developer workflow." For teams using Claude Code, Cursor, or Copilot, this closes a long-standing gap — agents can now iterate on design and implementation in the same loop.

**Cisco ships DefenseClaw at RSA 2026 — Zero Trust comes to AI agents** — Cisco unveiled DefenseClaw at RSA Conference, an open-source security framework for agentic AI that ships to GitHub on March 27th. It bundles five tools: a skills scanner, MCP scanner, agent-to-agent scanner, CodeGuard static analysis, and an AI bill-of-materials generator. The headline feature is Zero Trust Access for AI agents — Cisco extended its Duo IAM platform to give agents registered identities, map them to human owners, define time-bound permissions, and enforce MCP-based policy controls in its Secure Access SSE product. This is the first major enterprise security vendor to ship a purpose-built, comprehensive framework for agentic AI governance, including MCP enforcement at the network edge.

**Artemis II on track for April 1st** — A quick heads-up for the week ahead: NASA's Artemis II mission is targeting launch no earlier than April 1st, with a window running through April 6th. SLS and Orion are on Pad 39B at Kennedy Space Center. This is the first crewed deep-space mission since Apollo 17, more than fifty years ago. NASA says no technical issues are outstanding. The crew of four, including Canadian astronaut Jeremy Hansen, will fly a lunar flyby and return — no landing — but it's a significant milestone for human spaceflight regardless.

That's your agent platform briefing for Wednesday, March 25th. Big week ahead with the Anthropic ruling expected and Artemis on the pad.