CLAUDE·news
PTEN
№ 0612026-06-12

Claude Report — 2026-06-12

🚀 Anthropic Official

Claude Fable 5 + Claude Mythos 5 — Generally Available

June 9. claude-fable-5 is Anthropic's most capable widely released model, built for demanding reasoning and long-horizon agentic work; claude-mythos-5 (same capabilities, safety classifiers removed) is restricted to Project Glasswing participants only. Both share a 1M token context window, 128k max output tokens, always-on adaptive thinking, and pricing of $10/M input + $50/M output. Fable 5 is generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry; free on Pro/Max/Team/seat-based Enterprise through June 22, then moves to usage credits from June 23.

Fable 5 Integration Guide — Refusals, Tokenizer, Data Retention

June 9. Three breaking changes for integrations calling claude-fable-5: (1) safety classifiers can decline requests, returning stop_reason: "refusal" as HTTP 200 — not billed; the new beta fallbacks parameter lets the API retry on another model automatically, and a fallback credit refunds prompt-cache cost on retry; (2) the tokenizer produces ~30% more tokens for the same text vs. pre-Opus 4.7 models — use the token-counting API with model: "claude-fable-5" to remeasure; (3) mandatory 30-day data retention overrides zero-retention enterprise agreements (data is not used for training). Raw thinking is never returned; thinking.display: "summarized" | "omitted" controls what thinking blocks contain; the effort parameter controls reasoning depth.

Claude Code v2.1.175 — enforceAvailableModels Managed Setting

June 12. Adds enforceAvailableModels managed setting: when enabled, the availableModels allowlist also constrains the Default model (a Default resolving to a disallowed model falls back to the first allowed model), and user or project settings can no longer widen a managed availableModels list.

Claude Code v2.1.174 — VSCode Usage Attribution + 10+ Fixes

June 12. Adds wheelScrollAccelerationEnabled setting to disable scroll acceleration in fullscreen; [VSCode] adds usage attribution to the Account & usage dialog, breaking out cache misses, long context, subagents, and per-skill/agent/plugin/MCP consumption. Fixes: /model picker hiding which model Default resolves to, Fable 5 credit banner incorrectly shown for enterprise usage-based-billing accounts, Bedrock GovCloud (us-gov-*) deriving the wrong inference-profile prefix, background sessions inheriting another session's ANTHROPIC_* provider env vars, a 1–2 second exit pause on macOS/Linux, wrong model names in git co-author attribution, /advisor dialog pre-selecting a model blocked by availableModels, skill hot-reload re-sending the full skill listing, Workflow tool subagents missing attribution headers, and pre-warmed background worker authentication failures.

Claude Code v2.1.170 — Fable 5 Added to Claude Code

June 9. Introduces claude-fable-5 as a selectable model in Claude Code; fixes sessions not saving transcripts when launched from the VS Code integrated terminal or from shells with inherited Claude Code environment variables.

Claude Fable 5 Generally Available in GitHub Copilot

June 9. Available to Copilot Pro+, Max, Business, and Enterprise across VS Code (chat/ask/edit/agent modes), Visual Studio, Copilot CLI, Copilot cloud agent, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse; rollout is gradual. Enterprise and Business admins must explicitly enable the Fable 5 policy in Copilot settings (disabled by default); unlike other Copilot models, Fable 5 requires mandatory 30-day data retention.

API: Self-Hosted Sandbox Pending-Work Endpoint on Claude Platform on AWS

June 10. The GET /v1/environments/{id}/work endpoint, which lists pending work for self-hosted sandboxes, is now available on Claude Platform on AWS; requires the GetEnvironment IAM action.

Claude Corps — National AI Fellowship for Nonprofits

June 11. Anthropic launches Claude Corps, placing 1,000 early-career workers (under 2 years full-time experience, US work-authorized) at nonprofits for one year at $85k salary plus benefits; fellows receive intensive Claude training upfront and five hours of weekly skill development during placement. First cohort applications close July 17; program begins October 2026.

🔌 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 v0.3.174 — model_fallback Event Expanded to All Triggers

June 12. SDK consumers now receive the system/model_fallback message for all fallback triggers — overloaded, server_error, and last_resort in addition to the previously covered model_not_found and permission_denied; the message's trigger field gains the server_error and last_resort values.

claude-agent-sdk-typescript v0.3.170 — Fable 5 Model Types Added

June 9. Adds the claude-fable-5 model constant and the fable alias to SDK model type definitions, alongside parity with Claude Code v2.1.170.

Unchanged in window: nicobailon/pi-subagents, Chachamaru127/claude-code-harness

🔗 MCPs & Integrations

claude-code-action v1.0.146 — Image Detection, Auth Fallback, SDK Fix

June 12. Fixes image-type detection (now reads magic bytes instead of URL extension), SDK iterator hang in pull_request runs, fallback to inherited env for auth when inputs are empty, and stale claude-prompts dir written before each run; adds labels to formatContext() output for issues and PRs; 9 new contributors.

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

💡 Community — Workflows & Ideas

HN: "If Claude Fable stops helping you, you'll never know"

~June 9-12. Fable 5's classifiers return HTTP 200 with stop_reason: "refusal" rather than an error code, meaning silent fallbacks to Opus 4.8 are easy to miss in production; thread catalogs patterns for detecting refusals via stop_reason field checks and flags monitoring dashboards that rely solely on HTTP status codes.

HN: "Anthropic requires 30-day data retention for Fable and Mythos"

~June 9-12. Covers the policy that Fable 5 and Mythos 5 impose mandatory 30-day retention even on previously zero-retention enterprise agreements; discussion surfaces implications for regulated industries (finance, healthcare, legal) and data-residency compliance programs.

HN: "Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable"

~June 10-12. Mythos 5's cyberdefense capabilities are locked behind Project Glasswing; thread covers the friction this creates for security research workflows, the gap between Fable 5 and Mythos 5 in offensive-security tasks, and community-proposed mitigations.

📰 Quick Mentions

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

← back to archiveend of edition № 061