Seoul office opens, WIF replaces API keys, Claude Design hits 1M users

Anthropic officially opened its Seoul office on June 17 — its third in Asia-Pacific after Tokyo and Bengaluru — led by KiYoung Choi, a veteran of Snowflake, Google Cloud, Microsoft Korea, and Adobe. Alongside the opening, a wave of Korean enterprise deployments was announced: NAVER rolled out Claude Code across its entire engineering organization, Samsung SDS deployed Claude Cowork and Claude Code across Samsung Electronics, LG CNS is rolling out Claude to thousands of employees with plans to extend across LG Group, Nexon uses Claude Code for live-service game software, and Hanwha Solutions is deploying Claude globally via AWS Bedrock with in-region data residency controls. Korea now ranks in the top dozen countries globally for Claude.ai use, and APAC run-rate revenue has grown over 10x in the past year.

As part of the Seoul launch, Anthropic signed a Memorandum of Understanding with Korea's Ministry of Science and ICT to advance AI safety in the public sector. The two parties plan to conduct Korean-language model safety evaluations with the Korea AI Safety Institute and share information on AI-enabled cyber threats. Anthropic will also provide Claude access to up to 60 researchers at the National AI Research Lab (NAIRL) — a consortium of KAIST, Korea University, Yonsei University, and POSTECH — supporting work on AI safety, alignment, robustness, and model evaluation.

On the startup and nonprofit side: Channel Corp uses Claude to power Channel Talk, its customer AI platform used by over 230,000 businesses across Korea, Japan, and the US. Good Neighbors Korea — a child-rights NGO — is deploying Claude to help staff analyze program outcomes and navigate social welfare law, freeing workers from administrative tasks to focus on the field. Claude for Startups is now live in Korea, and Anthropic co-hosted a Claude Build Day with BASS Ventures, bringing together over 100 Korean founders and developers.

Workload Identity Federation is now generally available on the Claude Platform, replacing long-lived sk-ant-* API keys with short-lived OIDC tokens from identity providers you already operate: AWS IAM roles, GCP or Kubernetes service accounts, Azure Managed Identity, GitHub Actions tokens, Okta, SPIFFE, or any OIDC-compliant provider. Each workload now gets its own service account with its own identity, roles, and audit trail. Federation configuration is fully programmatic via new Admin API endpoints for issuers, service accounts, and federation rules. API keys still work alongside WIF, so you can migrate one workload at a time. To get started: Claude Console → Settings → Workload identity.

Admins can now provision MCP connectors for their entire organization through their identity provider, starting with Okta. Users get connector access automatically on first login, with authorization configured centrally — no per-user setup required. The beta covers Claude chat, Claude Code, and Cowork on Team and Enterprise plans. Previously, admins enabled a connector org-wide but every individual user still had to authorize it themselves.

The Claude Developer Platform shipped new tool versions for code execution, web search, and web fetch. The new code_execution_20260521 version now discloses the 90-second per-cell execution time limit in the tool description so Claude can budget long-running cells appropriately. A new response_inclusion field lets you trim consumed result blocks for agentic workflows — useful for keeping context windows lean in multi-step pipelines. No beta header required for either change.

Claude Code's auto mode received a significant safety upgrade: destructive git commands (git reset --hard, git checkout -- ., git clean -fd, git stash drop) are now blocked unless you explicitly asked to discard local work. git commit --amend is blocked if the commit wasn't made by the agent this session. terraform destroy, pulumi destroy, and cdk destroy are also blocked unless you explicitly named the stack to destroy. These guardrails ship alongside clearer model deprecation warnings on stderr in print mode, and a new /config help command listing every configurable key.

A new /config key=value syntax lets you set any Claude Code setting directly from the prompt — for example, /config thinking=false — without editing config files or restarting the session. Works in interactive mode, -p (print) mode, and Remote Control. Also added: sandbox.allowAppleEvents opt-in for sandboxed macOS commands, and a Bun runtime upgrade for faster startup.

Claude Code's auto mode — which replaces per-action permission prompts with background safety checks — is now available on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry for Opus 4.7 and Opus 4.8. Opt in by setting CLAUDE_CODE_ENABLE_AUTO_MODE=1 in your environment, then cycle to auto mode with Shift+Tab. Also in this release: Claude Code now prompts before writing files that can run code (shell startup files like .zshenv, git config, build-tool configs like .npmrc and .bazelrc) even in acceptEdits mode.

Claude Design — Anthropic's design-to-code tool — surpassed 1 million users in its first week after launch. The product received an update this week adding design system consistency across projects, tighter Claude Code sync, direct canvas editing, stronger layout controls, and more tool connections for smoother design-to-code handoff. Claude Design now lives in the sidebar of the Claude desktop app and is also accessible at claude.ai/design.

As announced at launch, Fable 5's inclusion in Pro, Max, Team, and seat-based Enterprise plans at no extra cost ends on June 22, 2026. From June 23, access to Fable 5 requires usage credits. Anthropic has said it plans to restore Fable 5 as a standard subscription feature as soon as capacity allows. Note: Fable 5 and Mythos 5 remain suspended for non-US citizens following the June 12 export control directive; US citizens can still access Fable 5 via usage credits from June 23.