CLAUDE·news
PTEN
№ 0622026-06-13

Claude Report — 2026-06-13

🚀 Anthropic Official

US Government Export Control Directive — Fable 5 and Mythos 5 Disabled Globally

June 12. The US Commerce Department issued an export control order requiring Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national (inside or outside the US, including Anthropic employees); because real-time nationality filtering is technically infeasible, Anthropic shut down both models for every customer worldwide at 9:59pm ET. Anthropic states the cited jailbreak — prompting the model to read a codebase and fix software flaws — is narrow, non-universal, and reproducible in other public models; it is complying while disputing the rationale and pledging to share technical details and restore access as soon as possible.

Fable 5 Silent Model-Switching Acknowledged — Visible Refusals Pledged

June 11. Anthropic acknowledges that Fable 5 was silently rerouting requests about AI model development, cybersecurity, chemistry, and biology to Opus 4.8 without notifying users; the company calls this "the wrong trade-off" and commits to replacing silent degradation with explicit, visible refusal messages going forward.

Claude Code v2.1.172 — Nested Sub-Agents and Plugin Marketplace Search

June 10. Introduces nested sub-agents — agents can now spawn additional sub-agents up to 5 levels deep; adds a search bar to the /plugin marketplace browser; Amazon Bedrock now reads the AWS region from ~/.aws config files when the environment variable is absent; fixes sessions on 1M-context plans getting permanently stuck without usage credits; reduces idle CPU usage during background agent execution.

Claude Code v2.1.173 — Fable 5 Model Name Normalization

June 11. Fixes Fable 5 model names containing a [1m] context-window suffix not being normalized (suffix is now stripped automatically); resolves a spurious "sandbox dependencies missing" startup warning on Windows.

Policy on the AI Exponential — Dario Amodei's Regulatory Framework

June 10. Amodei's essay argues AI capabilities are compounding faster than policy can respond and calls for a shift from transparency requirements to binding, FAA-style regulation: mandatory third-party testing of frontier models before deployment, government authority to block unsafe releases, secured model weights, and incident reporting with revenue-based civil penalties; the document accompanies a $350M commitment comprising a $200M Economic Futures Research Fund and a $150M fellowship for early-career workers.

Anthropic Public Record — First Wave Results

June 12. Inaugural survey of ~52,000 Americans (fielded Nov–Dec 2025): 64% fear job displacement, 56% fear cognitive dependency; only 15% trust AI companies to make development decisions — the lowest trust score of all institutions tested; 70%+ support government AI regulation across party lines; daily AI users are 16 percentage points less worried about job displacement than non-users.

TCS × Anthropic Partnership

June 12. Tata Consultancy Services deploys Claude across 50,000 employees in 56 countries and builds Claude-powered solutions for financial services, healthcare, and public-sector clients; Diligenta (TCS's UK life-and-pensions unit) applies Claude to 22M+ policyholders, and TCS iON delivers Claude training through 75M annual assessments in India.

🔌 Claude Code Plugins

Plugin cache unavailable — top-10 tracking skipped this run.

🛠️ Skills

None of the top 5 had verifiable updates in the window.

Unchanged in window: anthropics/knowledge-work-plugins, glebis/claude-skills, anthropics/skills, Piebald-AI/claude-code-system-prompts, ComposioHQ/awesome-claude-skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.176 — Turn Result Messages and Session State Fixes

June 12. Fixes turn result messages being dropped when multiple turns complete while a background agent or workflow is running; fixes background agent, remote agent, and MCP task state not being restored when resuming a session via the SDK.

Unchanged in window (parity-only releases): v0.3.177 (Jun 13), v0.3.175 (Jun 12), v0.3.173 (Jun 11); nicobailon/pi-subagents; Chachamaru127/claude-code-harness

🔗 MCPs & Integrations

Vercel AI SDK 7 — HarnessAgent with Claude Code Adapter

June 12. Introduces HarnessAgent, a unified API for running Claude Code, Codex, and Pi agent harnesses from a single interface; the @ai-sdk/harness-claude-code package wraps Claude Code sessions with support for custom tools, skills, and sandboxed workspaces and returns AI SDK-compatible streams; marked experimental.

Unchanged in window: modelcontextprotocol/servers, modelcontextprotocol/typescript-sdk, awslabs/mcp, portainer/portainer-mcp, github/github-mcp-server

💡 Community — Workflows & Ideas

HN: "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

Jun 12–13. Top comments draw parallels to 1990s encryption export controls, argue the directive inadvertently advantages Chinese open-weight AI labs, and challenge whether prompting a model to review code for vulnerabilities constitutes a meaningful security threshold given comparable capability in GPT-5.5 and other public systems.

HN: "Anthropic apologizes for invisible Claude Fable guardrails"

Jun 11. 500+ point thread documenting Fable 5's silent model-downgrade behavior; top comments argue the AI-model-development guardrail is anti-competitive rather than safety-motivated and flag healthcare and research workflows where invisible output degradation creates real-world risk without any error signal to the developer.

📰 Quick Mentions

Relevant links that don't warrant a section of their own:

← back to archiveend of edition № 062