Fable 5 Lands and Then Gets Pulled, Plus Deeper Sub-Agent Trees

I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of June fifteenth. Claude Fable 5 arrived and then got pulled after a US export directive. Sub-agents can now spawn their own sub-agents. And a new safe mode flag isolates config problems.

Let's start with Claude Code -- the headline this week is a model that showed up, and then disappeared. Version two point one point one seventy added access to Claude Fable 5, a Mythos-class model that Anthropic positioned as its strongest generally available release. Three days later, a US government export directive forced Anthropic to suspend access to both Fable 5 and Mythos 5. If you previously routed work to Fable 5, auto mode now falls back to the best available Opus model, and there's a new classifier fallback cookbook that walks through detecting safety-classifier blocks and routing to Opus four point eight instead.

We'll link both in the show notes. Beyond the model story, the structural change worth knowing about is that sub-agents can now spawn their own sub-agents, up to five levels deep. That lets you decompose work into deeper trees without orchestrating everything from your top-level session -- a sub-agent investigating a bug can fan out its own helpers without coming back to ask you. Building on that theme of giving you more control, there's also a new safe mode flag.

Launching with safe mode disables your project instructions, plugins, skills, hooks, and MCP servers all at once, which is exactly what you want when you're trying to figure out whether a problem is coming from your configuration or from Claude Code itself. The same week brought a useful quality-of-life addition for anyone juggling multiple repos: you can now move an active session into a different working directory without invalidating the prompt cache. Pivot to another repo or worktree mid-conversation, and you don't pay the cache-miss tax. And for administrators, there's a new managed setting that makes your available models allowlist also bind the default model, and prevents user or project settings from widening a managed list.

If a default would resolve to a disallowed model, it now falls back to the first allowed one. Given the Fable 5 routing changes, this is a good week to double-check those managed settings if you run a team. Rounding out the user-facing changes, the Visual Studio Code usage dialog now breaks down consumption by cache misses, long context, sub-agents, and per-skill, agent, plugin, and MCP usage over the last twenty-four hours or seven days. Session titles also now generate in whatever language you're actually working in.

On the under-the-hood side, the theme this week is lower CPU use and cleaner model routing. Long conversations now skip redundant message normalization and full history transforms when streaming tool-use state hasn't changed. The goal chip no longer redraws the terminal five times a second while idle. And there's a thirty to fifty millisecond per-turn UI stall that macOS claude dot ai users were hitting -- that's gone.

The model picker also got a careful pass. It now correctly shows the family that default resolves to, stops offering models that Bedrock doesn't actually serve, and respects your available models list for sub-agent overrides, the advisor model, and version-specific IDs. Bedrock now reads your AWS config files when the region environment variable isn't set, and GovCloud regions no longer derive a bad inference profile prefix.

Over to broader Anthropic news, and most of it traces back to the Fable 5 story. The New York Times is reporting on the export directive itself, framing it as the US barring foreigners from using Anthropic's most advanced models. The Wall Street Journal has reporting that connects the dots on how it happened -- according to the Journal, conversations between Amazon's CEO and US officials helped trigger the crackdown. That's useful backstory on how cloud-provider politics are shaping which models actually reach you.

Sitting alongside that policy story is a safety story from Axios, which covers Anthropic's own evaluation finding that Mythos can convert software patches into working exploits within hours. That evaluation was part of the safety case that preceded both the Fable 5 launch and the subsequent US restriction. And separately, the Wall Street Journal also has a piece on intensifying price competition between Anthropic and OpenAI, with OpenAI reportedly considering steep cuts. Worth tracking if model pricing affects how you size your Claude Code usage.

We'll link all of these in the show notes. That's it for the brief. I'm Shannon, and we'll see you next week.