US Greenlights Limited Mythos 5 Re-Release for Critical-Infrastructure Defenders
The US government has cleared Anthropic to redeploy Mythos 5 — its most capable cybersecurity model — to a vetted set of 100+ US companies and federal agencies that operate and defend critical infrastructure. Mythos was pulled offline June 12 under an export-control directive after Amazon researchers showed a jailbreak that bypassed its cyber classifier by framing requests as defensive code review.
The carve-out is narrow and defensive-only: access is restored just to organizations the government has named, so they can use the model to hunt vulnerabilities and harden their own systems. Notably, the letter does not cover Fable 5, the less powerful consumer-facing sibling, which remains suspended on day 18. The frontier is reopening — but through a government gate, one approved org at a time.
Claude Launches Generally Available in Microsoft Foundry on Azure
Claude is now generally available in Microsoft Foundry, giving Azure customers native access with the identity, billing, and governance they already run — plus optional US data zone support. Claude Opus 4.8 and Claude Haiku 4.5 are live in the Messages API for coding, agentic work, and complex reasoning.
This is the multi-cloud play maturing: Claude already lived on AWS Bedrock and Google Vertex, and a first-class Azure path removes the last “but we’re a Microsoft shop” objection. For enterprises standardizing their stack, Claude is now reachable wherever the data and the compliance boundary already sit.
Managed Agents Get Self-Hosted Sandboxes and Private MCP Servers
Claude Managed Agents can now run in a sandbox you control and connect to your private Model Context Protocol (MCP) servers. With self-hosted sandboxes in public beta, tool execution moves to an environment you configure — your own infrastructure or a managed provider like Cloudflare, Daytona, Modal, or Vercel — while the agent loop stays on Anthropic’s side.
That split is the whole point: you keep code execution and sensitive data inside your perimeter, Anthropic keeps running the reasoning. For regulated teams that wanted agents but couldn’t ship data to a black-box sandbox, this is the architecture that unblocks a pilot.
Claude Code Adds Nested Sub-Agents, a Tool Marketplace, and Per-Agent Cost Tracking
The June Claude Code wave is heavy on agent plumbing: nested sub-agents up to three levels deep for layered task decomposition, a fallbackModel setting for resilient model chains, and a community tool marketplace for sharing and discovering tools. Scoped permissions enforce least privilege on sub-agents, and usage attribution breaks cost down per agent and per task.
Week 26 also shipped claude mcp login and claude mcp logout for authenticating MCP servers from your shell, shell mode that reacts to output without a second prompt, and a roughly 37% cut in CPU use during streaming. The per-task cost breakdown is the quiet headline — after Uber’s budget blowout, finance-grade attribution is exactly what teams were asking for.
Claude Design Crosses 1M Users; Excel and PowerPoint Add-Ins Share Context
Claude Design drew more than one million users in its first week and just got stickier: it now holds to your design system across projects, works fluidly alongside Claude Code, supports direct on-canvas editing, and plugs into more of the tools you already use. The pitch is a design surface that respects your existing system instead of reinventing it each session.
Separately, the Claude for Excel and PowerPoint add-ins now share full conversation context, so an action Claude takes in one app is informed by everything that happened in the other. Build the model in Excel, and the deck in PowerPoint already knows the numbers — the Office workflow is starting to feel like one continuous session.
The Mythos Saga Is the Real Story of Anthropic’s IPO Year
The numbers are staggering on their own: a $65B Series H at a $965B post-money valuation, a confidential IPO filing on June 1, run-rate revenue past $47B, and a guide to $10.9B in Q2 revenue with a first profitable quarter in sight. By that scoreboard, Anthropic is winning. But the more telling event this month wasn’t the financing — it was watching the US government reach in and switch off two frontier models for 18 days and counting.
That’s the trade Anthropic is making real-time as a public-company-in-waiting: the same capability that earns a near-trillion-dollar valuation also makes the frontier a national-security asset the state will gate. Mythos coming back for critical-infrastructure defenders, while Fable stays dark, sketches the shape of the future — the strongest models reaching vetted institutions first, consumers second. For investors reading the S-1, the durable question isn’t revenue growth. It’s who actually controls the release valve on the product.