🧭 US Government Export Control Directive Suspends All Access to Fable 5 and Mythos 5

On June 12 at 5:21pm ET, Anthropic received a US government export control directive requiring the immediate suspension of access to Claude Fable 5 and Claude Mythos 5 for all users worldwide. The directive, which Anthropic describes as citing national security authority, was triggered by the government's claim that it had identified a method for bypassing the models' safety restrictions — a "jailbreak." Anthropic has complied with the directive while publicly disputing its proportionality, and has filed legal challenges. Claude Code v2.1.177, which shipped the same day, automatically falls back to Claude Opus 4.8 whenever Fable 5 is selected — no configuration change is needed on your end.

What Anthropic says happened

• The government demonstrated a specific jailbreak technique to Anthropic, which identified it as targeting a narrow set of previously known, minor vulnerabilities.
• Anthropic reviewed the technique and disagrees that a narrow potential jailbreak — of the kind that exists across all frontier models — warrants recalling a commercial model deployed to hundreds of millions of users.
• Anthropic has stated: "If this standard was applied across the industry, it would essentially halt all new model deployments for all frontier model providers."
• The suspension covers all access globally — not only foreign nationals — because Anthropic could not rapidly implement geography-gated enforcement without blocking access across its entire platform.

What this means for developers

All other models remain fully available: Opus 4.8, Sonnet 4.6, Haiku 4.5, and all older model versions continue to operate normally. If your production code pins to claude-fable-5-20261009 or claude-mythos-5-20261009, requests will currently fail at the model-selection layer. The recommended immediate action depends on your environment:

// If you use ANTHROPIC_MODEL or model IDs in code, update to:
const model = "claude-opus-4-8-20260530";   // best currently available

// In settings.json (Claude Code), if you had Fable 5 pinned:
{
  "model": "claude-opus-4-8-20260530"   // explicit pin — do not rely on alias fallback
}

// Claude Code v2.1.177 handles the fallback automatically if you had
// Fable 5 selected interactively — but hard-coded API clients need a manual fix.

🧭 DXC Technology and TCS Join Claude Partner Network as Global Premier Partners — on the Same Day

On June 11, Anthropic announced two separate multi-year partnerships with DXC Technology and Tata Consultancy Services (TCS), both elevated immediately to Global Premier status in the Claude Partner Network. Together they bring 165,000+ engineers across 70+ countries into the Claude ecosystem and signal a significant push into the world's most demanding enterprise environments: banks, airlines, insurers, government agencies, and regulated healthcare.

DXC Technology — Claude inside mission-critical managed services

DXC (115,000 employees, operations in 70 countries) is one of the world's largest IT managed-services firms. The partnership is built on a concrete track record: DXC used Claude to build DXC OASIS, its AI-native orchestration platform for managed services, achieving an estimated 10× acceleration in software delivery with more than 95% of code generated by Claude before human review. DXC OASIS, launched April 2026, is now deployed across 50+ enterprise customers and will roll out across DXC's full global customer base under this alliance. As a Global Premier Partner, DXC will train tens of thousands of Claude-certified engineers who will embed Claude directly into the mission-critical infrastructure stacks that DXC operates for banks, airlines, manufacturers, and governments.

TCS — 50,000 engineers trained for regulated industries

TCS (600,000+ employees globally) enters the partnership as "customer zero" — training an initial cohort of 50,000 engineers across engineering, finance, legal, marketing, and sales before scaling Claude-based solutions to clients. The focus is explicitly on regulated industries where AI adoption has historically stalled at the pilot stage: financial services, public services, life sciences, healthcare, aviation, telecom, and medtech. TCS plans a dedicated Claude practice unit with consultants, engineers, and industry specialists who design and run Claude-based systems under the strict accuracy and auditability standards these sectors require.

🧭 June 15 Final Countdown: Agent SDK Billing Split and Model Retirements Take Effect Tonight

Two major platform changes that were announced on June 1 take effect at midnight — tonight. If you have not yet acted on either, this is the final window. The changes are: (1) the Agent SDK workload credit pool separates from subscription billing, and (2) Claude Sonnet 4 and Opus 4 base models are formally retired.

Change 1: Agent SDK credit pool separation

Starting tonight, all programmatic usage via the Agent SDK, the claude -p command, Claude Code GitHub Actions, and third-party agent applications will be billed against a dedicated, separately tracked Agent SDK Credit pool — no longer from the same pool as interactive Claude.ai chat:

• Pro plan: $20/month credit pool at standard API list rates
• Max 5× plan: $100/month credit pool
• Max 20× plan: $200/month credit pool
• Credits are per-user (not pooled across a team), refresh with your monthly billing cycle, and do not roll over
• Once depleted, automated requests stop entirely unless overflow billing is manually enabled

Change 2: Sonnet 4 and Opus 4 base model retirements

The original claude-sonnet-4-20260514 and claude-opus-4-20260514 base model IDs are retired tonight. Any API call or settings.json entry still referencing these specific model IDs will receive an error response. Acceptable replacements:

# Retired tonight — will return model-not-found errors after midnight:
claude-sonnet-4-20260514
claude-opus-4-20260514

# Replacements:
claude-sonnet-4-6-20260714   # Sonnet 4.6 — drop-in improvement
claude-opus-4-8-20260530     # Opus 4.8 — drop-in improvement
                             # (Fable 5 is currently suspended — see Entry 1)

Five-minute pre-midnight checklist

• Grep your codebase and settings.json files for claude-sonnet-4-20260514 and claude-opus-4-20260514 — update both.
• Open Claude.ai → Account → Billing and confirm your Agent SDK credit pool appears and overflow billing is configured to your preference.
• Check any CI/CD pipelines or cron jobs that invoke claude -p or Claude Code GitHub Actions — these will stall if the credit pool is empty and overflow is off.
• If you run Managed Agents scheduled tasks, verify that the schedule cadence won't exceed your credit pool within the first billing cycle.
• Review any shared-team workflows: credits are per-user, not per-team, so a high-volume team member's runs cannot draw from a colleague's credit pool.