US Lifts Mythos 5 Block for a Named List of Trusted Partners
The Commerce Department cleared Anthropic to release Claude Mythos 5 to roughly 100 US companies and federal agencies, a major de-escalation of the standoff that began when export controls took the model offline on June 12. Commerce Secretary Howard Lutnick wrote to Anthropic that “appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.”
Under the arrangement, no license is required to export the model to the entities named in the letter’s Annex A and their foreign-national employees. Lutnick noted he can amend that allowlist — and re-impose requirements — “at any time.” The thaw reportedly followed Anthropic dispatching a team of senior scientists and engineers to work directly with Commerce and the Office of the National Cyber Director.
Alibaba Distillation Fight Heads to Congress
Anthropic told US senators that operators tied to Alibaba’s Qwen lab used roughly 25,000 fraudulent accounts to run nearly 29 million exchanges with Claude between April 22 and June 5, targeting software engineering and agentic reasoning — the model’s most commercially valuable skills. Anthropic calls it the largest distillation campaign it has identified, after naming DeepSeek, Moonshot and MiniMax in February.
The accusation is already moving policy: Senators Bill Hagerty and Andy Kim plan an amendment to must-pass defense legislation that would blacklist or sanction Chinese firms found improperly accessing US model output. Distillation lets a rival train a cheaper clone from a frontier model’s answers — often without the safety guardrails of the original.
Claude Code Adds Shell-Level MCP Login and a Sturdier /rewind
This week’s Claude Code drop is all about workflow friction. New claude mcp login and claude mcp logout commands let you authenticate or clear credentials for configured MCP servers straight from the shell, instead of digging through the interactive /mcp menu. Shell mode now responds to command output without a second prompt, and /rewind can resume a conversation from before a /clear wiped it.
The standout fix: background subagents now surface permission prompts in the main session rather than silently auto-denying them. For anyone running multi-agent jobs, that’s the difference between a stalled run and one that just asks for the access it needs.
Claude Lands as an Agent Provider Inside JetBrains IDEs
GitHub shipped a public preview that lets you pick Claude as an agent provider in JetBrains IDEs without leaving the editor: install the Claude Code CLI, point Copilot Chat’s settings at it, and select Claude from the agent picker. It’s part of a broader release that also adds org and enterprise agents.
One caveat worth flagging: in this preview the Claude agent runs in bypass-permissions mode, so file edits and tool calls are auto-approved — configurable permissions are promised in a later release. Useful, but not where you’d point it at anything irreversible yet.
Claude’s Paying Consumer Base Is Up About 75% This Year
Credit-card transaction data from analytics firm Indagari shows Anthropic’s paying consumers and revenue climbing month over month — up roughly 75% since January. Claude’s global market share rose from 5.1% in February to about 10% in April, while ChatGPT’s share slipped below 50% for the first time, down from 81% two years ago.
The signal isn’t just spend. “Claude” is now the most-searched term on DataCamp’s site — ahead of “AI” itself — and demand for Claude courses is outpacing ChatGPT three to one among self-directed learners. ChatGPT still leads by a wide margin, but the gap is narrowing in the segment that pays.
Fable 5 Stays Blocked — Consumers Still in Limbo
The Mythos 5 clearance pointedly did not cover Fable 5, the weaker, consumer-facing model that was briefly the most powerful AI widely available to the public before June 12. The Commerce letter is silent on it. People close to the talks say a Fable release is the direction of travel, but the timeline is unclear.
The irony is that Fable carries stronger guardrails than Mythos — it’s blocked from answering a range of cyber and bio questions — yet it’s the one still offline for everyday users. For now, trusted enterprises get the frontier model back while consumers keep waiting on Opus 4.8.
Frontier AI by Allowlist Is a New Kind of Distribution
The Mythos 5 deal is more interesting than “ban lifted.” What Commerce actually created is a government-maintained allowlist — a named set of trusted partners who can run the most powerful model, with the secretary holding the power to add or drop names “at any time.” That turns frontier-model access into something closer to a security clearance than a product launch, and it lands the same week Anthropic is asking Congress to sanction Chinese labs for distilling Claude.
Read the two stories together and a picture forms: capability is increasingly gated, not by price or rate limits, but by who you are. Enterprises on the list get Mythos; consumers wait on Fable; adversaries get accused of stealing what they can’t buy. For everyone building on Claude, the takeaway is that access tiers are becoming a policy variable, not just a billing one — and that’s a harder thing to plan a roadmap around.