US Lifts Export Controls on Fable 5 and Mythos 5; Anthropic Redeploys Globally
The US Department of Commerce notified Anthropic that it has removed the export restrictions on Claude Fable 5 and Mythos 5 — the two frontier models that had been suspended or geofenced for most users. Anthropic responded by redeploying Fable 5 globally and beginning to restore Mythos 5 access, pairing the rollout with deeper cyber safeguards, a draft AI jailbreak severity framework, and a HackerOne program for security researchers to report potential cyber jailbreaks.
It’s a notable reversal. For weeks the story was about restriction — a suspended model, a paid meter, limited access. Now the constraint is off, and Anthropic is racing to get its most capable models back in front of everyone while layering on the safety scaffolding that likely helped clear the path in the first place. Capability and governability, arriving together.
PwC Deepens Its Claude Alliance: A Joint Center of Excellence and 30,000 Certifications
Anthropic and PwC announced an expansion of their strategic alliance, deepening how PwC uses Claude to build technology, execute deals, and reinvent enterprise functions for clients across every industry it serves. The two are standing up a joint Center of Excellence and a program to train and certify 30,000 PwC professionals on Claude.
This is the Partner Network thesis in action. A day after Anthropic put $100M behind its integrator program, the marquee example shows why: the bottleneck on enterprise AI isn’t the model, it’s the army of consultants who actually stand it up inside a Fortune 500. Certify 30,000 of them and every one becomes a distribution channel.
Claude Ships Inside Apple’s Foundation Models Framework
Anthropic released ClaudeForFoundationModels, a Swift package that wires Claude into Apple’s Foundation Models framework across iOS 27, iPadOS 27, macOS 27, visionOS 27, and watchOS 27. It conforms Claude to the framework’s LanguageModel protocol, so you drive it with the exact same LanguageModelSession API you already use for Apple’s on-device model — respond(to:), streaming, guided generation, and tool calling all work unchanged.
The pattern is elegant: run cheap, private work on Apple’s on-device model, then hand off to Claude when a request needs multi-step reasoning, code generation, web search, or code execution. Requests go straight from the app to the Claude API — Apple never sees the prompts or responses. It targets the OS 27 betas, so APIs may still shift before GA, but for Swift developers this collapses the gap between on-device and frontier intelligence to a single session call.
MCP Tunnels and Self-Hosted Sandboxes Reach Preview
Two useful drops for agent builders. MCP tunnels arrived as a Research Preview, letting you connect Claude to MCP servers running inside your private network without exposing them publicly — a real unlock for teams whose data and tools live behind the firewall. Alongside it, self-hosted sandboxes for Claude Managed Agents let you run tool execution on your own infrastructure instead of Anthropic’s.
Anthropic also shipped connector observability in public beta plus in-app directory submission, so admins can monitor adoption, errors, latency, and usage across Claude products and submit MCP connectors directly from the app. The pitch to enterprises is consistent: bring Claude to where your infrastructure already is, and watch what it does once it’s there.
NVIDIA’s BioNeMo Agent Toolkit Plugs Into Claude Science
Fresh off last week’s Claude Science launch, Anthropic and NVIDIA connected the BioNeMo Agent Toolkit natively into the workbench, exposing GPU-accelerated life-sciences models as callable skills inside the Claude environment. A researcher can describe a task in plain language — analyze a genomic sequence, predict a protein structure, design a molecular binder — and Claude Science orchestrates the run against preconfigured, domain-specialized agents backed by NVIDIA’s accelerated microservices.
What makes it more than a demo is the design: BioNeMo stays open and harness-agnostic, so the same scientific skills work across different agent frameworks and independent research platforms, not just Claude. It’s the clearest sign yet that the “agents calling specialized tools” model is where real scientific compute is heading — with Claude positioning itself as the natural-language front door.
Last Week Was About Control. This Week Is About Reach.
If the theme running through late June was governability — spend alerts, entitlements, auditable artifacts — the story turning over this weekend is distribution. Look at where Claude is suddenly showing up: on your iPhone via Apple’s Foundation Models framework, inside your private network via MCP tunnels, in the lab via NVIDIA’s BioNeMo, and back across borders now that export controls on Fable 5 and Mythos 5 have lifted. Four very different surfaces, one direction: everywhere.
That’s the payoff of the boring layer Anthropic spent the last cycle building. Once you’ve made a model a CFO can budget and a CISO can trust, the next move is to put it in front of every developer, every researcher, and every enterprise workflow you can reach — and let 30,000 freshly certified PwC consultants carry it the last mile. Capability got Claude noticed. Control made it deployable. Reach is how it compounds.