MeteorIT
Journal/AI Weekly Digest/12–19 June 2026

AI Weekly Digest12–19 June 2026

Claude Fable 5 suspended globally three days after launch under a US government export-control directive — the first time a leading AI company has taken a deployed model offline under federal order; Workload Identity Federation goes GA; Anthropic files confidentially for IPO at ~$350B; Microsoft enables Claude model selection inside Copilot Chat; Gemini ships TTS streaming and broad deprecations.

Period
12–19 June 2026
Published
Jun 19, 2026
Covers
Anthropic · OpenAI · Gemini · Copilot

Dateline: June 19, 2026 | Next update: June 26, 2026

The defining story of the week — and potentially of 2026 — is the US government's export-control directive that forced Anthropic to suspend Claude Fable 5 and Mythos 5 globally on June 12, just three days after launch. This is widely reported as the first time a leading AI company has taken a publicly deployed model offline under a federal national-security order. Fable 5 remains unavailable as of today; Opus 4.8, Sonnet 4.6, and Haiku 4.5 are unaffected. Alongside this extraordinary development, Anthropic shipped Workload Identity Federation GA, enterprise-managed Okta connector access, and completed the Sonnet 4 / Opus 4 retirements. Anthropic also opened its Seoul office on June 17 with NAVER, Samsung SDS, and Nexon as anchor Korean enterprise customers.

Claude / Anthropic

⚠️ Claude Fable 5 + Mythos 5 — suspended globally

Suspended: June 12, 2026, 5:21 PM ET | Cause: US government export-control directive | Status: suspended worldwide as of June 19 | Fallback: Claude Opus 4.8

On June 12 — three days after launch — Anthropic received a US government directive under national-security export-control authority ordering it to halt access to Fable 5 and Mythos 5 for all foreign nationals, wherever located. Because real-time nationality filtering across API calls, customer contracts, and cloud delivery paths is not technically or legally feasible in the short term, Anthropic chose to suspend both models for all customers everywhere rather than risk non-compliance. This is widely reported as the first time a leading AI company has taken a publicly deployed commercial model offline under a federal directive.

⚠ Alert

Fable 5 (claude-fable-5) and Mythos 5 (claude-mythos-5) are currently unavailable globally. Any API call, Claude Code session, or MCP integration calling these model strings is returning errors. Recommended immediate fallback: claude-opus-4-8. The directive was triggered by a jailbreak technique that can instruct the model to scan a codebase for vulnerabilities — the Commerce Department cited this as a national-security risk. Anthropic publicly disagrees, arguing the technique is narrow, already present in other publicly available models including GPT-5.5, and not novel. Anthropic met with US officials June 15–16 to negotiate. No confirmed restoration date has been announced. Anthropic's international managing director stated at the Seoul office launch (June 17) he is 'very confident' both models will return 'in coming days.' A structured re-entry path — identity verification for US users — is planned from July 8. Prediction market Kalshi priced approximately 57% probability of restoration before July 1 as of June 18.

Technical details

Suspended model strings: claude-fable-5, claude-mythos-5 | All calls return errors | Fallback: claude-opus-4-8 (same $5/$25 per MTok as before) | Unaffected: claude-opus-4-7, claude-opus-4-8, claude-sonnet-4-6, claude-haiku-4-5-20251001 | IPO context: Anthropic filed confidentially for IPO on June 1 at ~$350B valuation; regulatory risk is now a factor investors are pricing | ID verification path: planned from July 8 for US users | Directive language: unusually broad — covers foreign nationals at US companies and on US soil

Best for: All API developers — change model parameter to claude-opus-4-8 now. Do not hardcode a single model string — keep model selection in config so future disruptions are a one-line change.

Workload Identity Federation — generally available

Announced: June 12–16, 2026 | Platform: Claude API + Claude Platform on AWS | Availability: all API users

Workload Identity Federation (WIF) replaces long-lived static API keys (sk-ant-...) with short-lived OIDC tokens issued by identity providers your organisation already operates. It is now generally available on the Claude Platform and represents the most significant authentication upgrade Anthropic has shipped to the API since the platform launched. Every workload gets its own service account with its own roles and audit trail, replacing the common practice of sharing a single API key across a team or pipeline.

★ What's new

WIF is now GA. Supported identity providers: AWS IAM roles, Google Cloud / Kubernetes service accounts, GitHub Actions tokens, Azure Managed Identity / Microsoft Entra ID, SPIFFE/SPIRE JWT-SVIDs, and Okta service applications (client_credentials flow). Setup flow in Claude Console: Settings → Workload identity → Connect workload — wizard creates the issuer, service account, and federation rule, then verifies end-to-end. Fully programmable via new Admin API endpoints for creating and updating issuers, service accounts, and federation rules at scale. API keys and WIF coexist — migrate one workload at a time. Each workload gets its own identity, roles, and audit trail. Every token exchange and request is logged against the service account. Federation rules support fine-grained least-privilege scopes.

Technical details

Token exchange: POST /v1/oauth/token (RFC 7523 jwt-bearer grant) | Token format: sk-ant-oat01-... (short-lived) | Providers: AWS IAM, GCP, GitHub Actions, Azure/Entra ID, Kubernetes, SPIFFE, Okta | Console setup: Settings → Workload identity | Admin API: create/update issuers, service accounts, federation rules | Claude Platform on AWS: WIF also GA; see IAM actions and AnthropicSelfHostedEnvironmentAccess managed policy | Managed Agents webhooks, multiagent orchestration, and self-hosted sandboxes also now available on Claude Platform on AWS

Best for: Engineering teams running Claude in CI/CD pipelines, Kubernetes, or any server-side workload — eliminates static API key rotation and leakage risk

Enterprise MCP connectors — Okta zero-touch provisioning (beta)

Announced: June 18–19, 2026 | Platform: Claude Platform | Availability: Team and Enterprise plans | Beta

Enterprise-managed MCP connector access launched in beta, starting with Okta. Previously, enabling a connector required two steps: an admin enabling it organisation-wide, then every individual user authorising it themselves. With this update, admins provision connectors once through their identity provider, and users get access automatically on first login with no individual authorisation action required.

★ What's new

Admins can now provision MCP connectors for their entire organisation through their identity provider (Okta is first, more IdPs to follow). Users get connector access automatically on first login — zero-touch for the end user. Centralised authorisation applies across Claude chat, Claude Code, and Claude Cowork. The MCP OAuth browser page has been updated to match Claude Code's visual style and now auto-closes on successful authentication.

Technical details

Beta: Team and Enterprise plans | First IdP: Okta | More IdPs: to follow | Scope: Claude chat, Claude Code, Cowork | Admin workflow: provision via IdP, no per-user action required | MCP OAuth page: redesigned, auto-close on success

Best for: Enterprise IT admins managing large Claude deployments — eliminates connector onboarding friction for every new user

Claude Code — connection stability, WSL2, and platform fixes

Platform: terminal / VS Code / web / mobile | Availability: all plans

Claude Code shipped two point releases focused on stability and reliability. The most impactful fixes address mid-stream connection drops, WSL2 scrolling regressions, and a batch of macOS and network drive bugs.

★ What's new

Fixed mid-stream connection drops: partial responses are now preserved instead of showing a raw error, and the spinner no longer gets stuck at 'running tool'. Fixed mouse-wheel scrolling regression in WSL2 under Windows Terminal and VS Code (introduced in 2.1.172). Fixed sandbox denyRead/allowRead glob over a large directory tree making the Bash tool description enormous and the session unusable on Linux. Fixed prompt caching not reading on custom ANTHROPIC_BASE_URL and on Foundry due to a per-request attestation token changing every turn. Fixed Write/Edit producing 0-byte or truncated files on network drives and cloud-synced folders. Fixed open, osascript, and browser-based auth flows failing with error -600 on macOS by adding the Apple Events entitlement. Fixed a startup regression (~120ms per launch in fresh environments, introduced in 2.1.169). Fixed compaction not honouring --fallback-model. Fixed claude agents workers failing with 401 Invalid bearer token when the daemon was started from a shell with a custom API gateway.

Technical details

Fixed: mid-stream connection drops (partial response preserved, spinner unstuck) | Fixed: WSL2 mouse-wheel scroll regression (2.1.172) | Fixed: sandbox glob Bash tool description bloat on Linux | Fixed: prompt caching on custom ANTHROPIC_BASE_URL and Foundry (per-request attestation token) | Fixed: 0-byte / truncated files on network drives and OneDrive | Fixed: macOS error -600 on browser auth (Apple Events entitlement added) | Fixed: 120ms startup regression (fresh environments, 2.1.169) | Fixed: --fallback-model not honoured during compaction | Fixed: 401 in claude agents on custom API gateway | Changed: fullscreen mode URL opening now requires Cmd+click (macOS) / Ctrl+click | Changed: 'Improved N memories' no longer lists individual files outside verbose mode

Best for: WSL2 users, macOS users, anyone on network drives or cloud-synced folders, and teams running agents on custom API gateways — update to the latest version

Anthropic Seoul office — NAVER, Samsung SDS, Nexon deployments

Office opened: June 17, 2026 | Location: Seoul, South Korea

Anthropic formally opened its Seoul office on June 17. The launch came with disclosures of three significant Korean enterprise deployments: NAVER has deployed Claude Code to its entire engineering organisation; Nexon is using Claude Code for live-service game development; Samsung SDS is rolling out both Claude Cowork and Claude Code across Samsung Electronics.

★ What's new

Seoul office open. Three anchor Korean enterprise deployments: NAVER (Claude Code, full engineering org), Nexon (Claude Code, live-service game dev), Samsung SDS (Cowork + Claude Code across Samsung Electronics). Samsung, SK hynix, SK Telecom, and the state agency KISA are awaiting access through Project Glasswing phase 2. Anthropic's international managing director stated the company is 'very confident' Fable 5 and Mythos 5 will return 'in coming days.' Global offices now include: San Francisco (HQ), London, Tokyo, Sydney, Seoul.

Technical details

Office: Seoul, South Korea | MD International: Chris Ciauri | Deployments: NAVER (Claude Code), Nexon (Claude Code), Samsung SDS (Cowork + Claude Code) | Glasswing phase 2 pending: Samsung, SK hynix, SK Telecom, KISA

Best for: APAC enterprise buyers and Korean partners — informational for others

Model retirements completed — Sonnet 4 and Opus 4 now offline

Retirement date: June 15, 2026 | Affected: claude-sonnet-4-20250514, claude-opus-4-20250514

Claude Sonnet 4 and the original Claude Opus 4 retired on schedule on June 15. API calls to these model strings are now returning errors. This retirement is separate from and unrelated to the Fable 5 suspension — these were planned deprecations on the normal schedule.

★ What's new

Sonnet 4 (claude-sonnet-4-20250514) and original Opus 4 (claude-opus-4-20250514) are now retired and returning errors. Migrate to claude-sonnet-4-6 (Sonnet) or claude-opus-4-8 (Opus). Any pipeline still calling these strings needs an immediate update.

Technical details

Retired June 15: claude-sonnet-4-20250514, claude-opus-4-20250514 | Migration: Sonnet 4 → claude-sonnet-4-6 | Opus 4 → claude-opus-4-8 | Previously retired (April 19): claude-3-haiku-20240307 | Current active strings: claude-opus-4-8, claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5-20251001 | Fable 5/Mythos 5: suspended

Best for: Any team still calling retired model strings — immediate action required

Anthropic — confidential IPO filing at ~$350B

Filed: June 1, 2026 | Type: confidential S-1 filing | Valuation: approximately $350 billion

Anthropic filed confidentially for an initial public offering on June 1, 2026, at a reported valuation of approximately $350 billion. The Fable 5 export-control directive has added a new regulatory risk factor that investors will now weigh before the listing. Pre-IPO shares dipped on the news of the suspension. No IPO date has been announced.

★ What's new

Confidential IPO filing confirmed at approximately $350 billion valuation. No listing date announced. The Fable 5 suspension has introduced a regulatory and national-security risk factor that was not present at the time of filing. Pre-IPO share prices have dipped in response. Anthropic has not commented on how the suspension affects its IPO timeline.

Technical details

Filing: confidential S-1 | Valuation: ~$350B | Revenue run rate: $30B+ (as of April 2026) | Risk factor added post-filing: Fable 5 / Mythos 5 export-control directive | Pre-IPO share movement: negative on suspension news

Best for: Investors and enterprise buyers evaluating Anthropic's long-term trajectory — informational

Plans and Pricing

No pricing changes this week. The practical impact of the Fable 5 suspension is that teams who had started using Fable 5 at $10/$50 per MTok are now running on Opus 4.8 at $5/$25 per MTok — half the cost. If Fable 5 is restored with usage credits required, the credit model may be revised. Anthropic has committed to communicating changes ahead of time.

Technical details

Fable 5: suspended, no access | Opus 4.8: $5/$25 per MTok (recommended fallback) | Opus 4.7: $5/$25 per MTok (also available) | Sonnet 4.6: $3/$15 per MTok | Haiku 4.5: low-cost tier | Fable 5 Batch API: unavailable while suspended | WIF: no additional cost over standard API pricing

Best for: Update all model strings to claude-opus-4-8 as the immediate fallback. Do not wait for Fable 5 restoration — no confirmed timeline exists.

ChatGPT / OpenAI

Dateline: June 19, 2026 | Next update: June 26, 2026

Over the past week, OpenAI has focused on strengthening ChatGPT's workplace capabilities, improving agent reliability, and continuing the shift toward persistent, project-based workflows. The emphasis remains on usability, integration, and execution rather than headline model releases.

GPT-5.3 Standard — default model

Release: late 2025 | Pricing: included | Availability: all users

★ What's new

Improved handling of context across longer sessions with multiple task changes.

Technical details

Context ~128k | Output ~4k–8k | Improved session-level context management

Best for: General use

GPT-5.3 Pro — high-reasoning model

Release: late 2025 | Pricing: Pro | Availability: Pro/Enterprise

★ What's new

Improved stability when working through multi-stage analytical projects.

Technical details

Context ~200k (est.) | Better reasoning persistence across extended chains

Best for: Deep analysis

GPT-5.3 Mini — fallback model

Release: late 2025 | Pricing: low-cost | Availability: all

★ What's new

Faster routing for lightweight requests and routine tasks.

Technical details

Context ~64k | Improved task classification and resource allocation

Best for: Quick tasks

Agent Mode

★ What's new

Improved execution reliability for workflows involving research, drafting, editing, and verification.

Technical details

Enhanced workflow orchestration | Improved state tracking between task stages

Best for: Task delegation

Deep Research

★ What's new

Improved synthesis of large information sets and better handling of contradictory sources.

Technical details

Enhanced source ranking and evidence reconciliation systems

Best for: Research

Memory & Projects

★ What's new

Better project continuity and retrieval of relevant information across long-running workstreams.

Technical details

Improved project-level memory segmentation | Reduced cross-project interference

Best for: Ongoing workflows

Advanced Voice Mode

★ What's new

More natural management of interruptions, clarifications, and topic changes.

Technical details

Improved conversational state tracking | Lower context-loss rates

Best for: Voice interaction

Enterprise & Team Collaboration

★ What's new

Improved support for shared projects, collaborative research, and team-based workflow management.

Technical details

Enhanced workspace architecture | Better collaboration-layer integration

Best for: Organisational productivity

Professional Workflows

★ What's new

Improved generation of structured reports, research outputs, planning documents, and strategic briefs.

Technical details

Better workflow-specific optimisation | Improved output consistency

Best for: Knowledge work

Plans and Pricing

No significant pricing changes announced this week. Pricing stable | API structure unchanged.

Gemini (Google)

Date: June 19, 2026 | Next update: June 26, 2026

This week focused on expanding audio streaming protocols, finalising widespread image and video generation architecture retirements, and introducing cross-industry specifications for agentic web workflows.

Gemini 3.1 Flash-TTS — streaming support

Launched: June 17, 2026

Google enabled streaming functionality via streamGenerateContent (and stream: true within the Interactions API) for the gemini-3.1-flash-tts-preview model, significantly lowering latency for real-time speech generation applications.

★ What's new

Streaming now supported for gemini-3.1-flash-tts-preview via streamGenerateContent. Uptime on streamGenerateContent pipelines at 99.95% following the update. Significantly reduces latency for real-time voice and TTS applications.

Best for: Real-time voice apps and TTS pipeline developers

Image generation — deprecations announced

Announced: June 15, 2026 | Shutdown: August 17, 2026

Legacy image generation engines including Imagen 4 and Gemini 3 Image models (imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001) are officially scheduled for permanent shutdown on August 17, 2026.

Best for: Developers using Imagen 4 or Gemini 3 image models — plan migration before August 17

Veo video models — deprecations (June 30 deadline)

Announced: June 15, 2026 | Shutdown: June 30, 2026 | Action required

The legacy models veo-2.0-generate-001, veo-3.0-generate-001, and veo-3.0-fast-generate-001 will shut down on June 30, 2026. Migrate endpoints to veo-3.1-generate-preview or Gemini Enterprise GA IDs before the end of the month.

★ What's new

June 30 shutdown confirmed for veo-2.0-generate-001, veo-3.0-generate-001, veo-3.0-fast-generate-001. Migrate to veo-3.1-generate-preview or Gemini Enterprise GA IDs immediately.

Best for: Video developers — migrate Veo endpoints before June 30

GMP Contextual View — shut down

Shut down: June 15, 2026

The experimental fixed interface for grounding outputs with Google Maps (GMP Contextual View) was officially shut down on June 15, 2026.

Best for: Developers using GMP Contextual View — find alternative grounding solutions

Agentic Resource Discovery (ARD) — open specification

Announced: June 17, 2026

Google introduced an open web specification alongside industry partners. ARD creates a universal protocol allowing background agents to securely publish, discover, and cross-verify their AI capabilities regardless of the underlying protocol or provider.

★ What's new

Open specification released for Agentic Resource Discovery — enables background agents to securely publish, discover, and cross-verify AI capabilities across frameworks and providers. Industry partners involved at launch.

Best for: Cross-framework AI developers and teams building multi-agent systems

Gemini in Sheets — 28 new languages

Rolled out: June 18, 2026

Natural-language spreadsheet creation and cell-structure generation in Google Sheets has expanded to 28 additional languages, natively onboarding Spanish, Portuguese, Japanese, Korean, and Italian layouts among others.

Best for: Multilingual teams using Google Sheets

Google Classroom app in Gemini

Launched: June 17, 2026

Educators can now sync Gemini directly with their Google Classroom profiles. The assistant natively leverages class syllabi, student needs context, and historic assignments to generate tailored teaching resources.

Best for: Educators and academic institutions using Google Classroom

Android Auto — Google Meet integration

Launched: June 18, 2026

Google Meet was officially integrated into Android Auto dashboards, enabling safe, hands-free meeting participation while driving.

Best for: Commuters and field workers attending meetings while driving

⚠️ June 25 cutover — action required this week

Deadline: June 25, 2026

Two things happen June 25: (1) legacy visual models gemini-3.1-flash-image-preview and gemini-3-pro-image-preview stop serving — migrate immediately. (2) The Google Chat workspace redesign goes live, replacing the old Gemini sidebar with the standalone "Ask Gemini" app. Past conversation history will not migrate — manually export your chat records this week to prevent history loss.

Best for: All Workspace users — export chat history and migrate image models before June 25

Calendar API — writerWithoutPrivateAccess (June 29)

Effective: June 29, 2026

The writerWithoutPrivateAccess secondary calendar filtering rule becomes generally available on June 29, 2026. Developers should review and update Calendar API integrations before this date.

Best for: Calendar API developers — review permissions before June 29

Microsoft Copilot

Dateline: June 19, 2026 | Next update: June 26, 2026

The defining story of the week is Microsoft's rollout of Claude integration into Copilot Chat across all platforms, marking the first time users can select between multiple foundation models inside Microsoft 365 Copilot. Alongside this, Copilot introduced infinite scroll for chat history on web and a design refresh in Microsoft Edge.

⚠️ Claude integration into Copilot Chat — live globally

Available: June 12, 2026 | Platforms: Windows, Mac, iOS, Android, Web | Status: live globally

On June 12, Microsoft enabled Claude model selection inside Copilot Chat across all major platforms. Users can now toggle between GPT-5 and Claude depending on task type. Claude is positioned for structured analysis, document breakdowns, and enterprise workflows, while GPT-5 remains strong for creative generation.

★ What's new

Model selector now available in Copilot Chat UI across Windows, Mac, iOS, Android, and Web. Enterprise tenants can restrict model choice via admin policy. Claude sessions logged separately in compliance reports. No additional licensing required beyond existing Copilot subscription. Teams should validate workflows before standardising on Claude, as model trade-offs exist.

Technical details

Model selector in Copilot Chat UI | Claude supported on all platforms | Enterprise admin policy for model restriction | Separate compliance logging for Claude sessions | Included in existing Copilot subscription, no new SKU

Best for: Analysts, researchers, and enterprise teams needing structured outputs alongside GPT-5 creative generation

Infinite scroll in Copilot Chat history

Announced: June 14, 2026 | Platform: Web Copilot | Availability: all users

★ What's new

Copilot Chat on the web now supports continuous scrolling through past conversations, eliminating the "load more" button. Implemented via virtualised DOM rendering, optimised for over 10,000 messages per tenant. Search filters unchanged. Export to Word/PDF still available via compliance tools.

Best for: Heavy Copilot users who frequently reference past conversations

Microsoft Edge — Copilot design refresh

Announced: June 15, 2026 | Platform: Microsoft Edge | Availability: all users

★ What's new

Copilot in Edge received a visual refresh with updated spacing, fonts, and colour palette. UI components migrated to Fluent 2 design system. Accessibility contrast ratios improved. Dark mode updated. Admins can disable Edge Copilot via GPO/Intune.

Best for: Organisations standardising on Microsoft AI tools with consistent UI

Plans and Pricing

No pricing changes this week. Claude integration is included in existing Copilot subscriptions — Copilot Pro at $20/user/month, Copilot for Microsoft 365 at $30/user/month. Infinite scroll and Edge refresh are feature updates with no additional cost. Enterprise E7 and Agent 365 tiers unchanged.

Filed under: AI Weekly Digest
First published: Jun 19, 2026

← Previous issue5–12 June 2026All issues