Fable 5 launches, then gets banned for foreign users by US export order — plus TCS, DXC, and Glasswing expand

Just three days after launch, the US Commerce Department — under Secretary Howard Lutnick — sent Anthropic a letter at 5:21pm ET ordering it to suspend all access to Fable 5 and Mythos 5 by any foreign national, inside or outside the US, including Anthropic's own foreign-national employees. Because Anthropic has no way to instantly verify every user's citizenship, it disabled both models entirely for all customers worldwide to ensure compliance. All other Claude models, including Opus 4.8, remain unaffected. Anthropic said the directive cited a 'potential narrow, non-universal jailbreak' that the government described only verbally, and that it believes the same technique would work against other public models too. This marks the first time a government has forced a leading AI lab to pull a publicly deployed frontier model offline.

Context matters here: in February 2026, contract negotiations between the Pentagon and Anthropic broke down after Anthropic refused to let Claude be used for lethal autonomous weapons or mass civilian surveillance. The DoD subsequently labeled Anthropic a 'supply chain risk' — a designation typically reserved for foreign adversaries — barring defense contractors from using Claude. Anthropic sued to overturn that designation, and the lawsuit is ongoing. Pentagon CIO Kirsten Davies posted on X that the department was 'prioritizing national security... Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation.' Anthropic says it believes the export order is a misunderstanding and is working to restore access.

A policy buried in Fable 5's system card revealed that the model would silently identify 'requests targeting frontier LLM development' and quietly limit its effectiveness — without telling the user. After a Wired report sparked major backlash, Anthropic apologized: 'We made the wrong tradeoff and we apologize for not getting the balance right.' Starting this week, any flagged request now visibly falls back to Opus 4.8 — the same pattern used for cyber and bio safeguards — and the API will return an explicit refusal reason for flagged requests (server-side fallback messaging coming within days). Anthropic said it chose invisible safeguards initially because visible ones are easier to probe and required more time to harden, but acknowledged that tradeoff was wrong.

Before the ban, Fable 5 launched as Anthropic's most capable public model ever: 80.3% on SWE-Bench Pro (vs 69.2% for Opus 4.8), first model over 90% on Hex's analytical benchmark, 1M token context by default, 128K max output, and always-on adaptive thinking. The Claude Developer Platform also added Managed Agents scheduled deployments, vault-based environment variable credentials, and richer session thread webhook events alongside the launch. Fable 5 was free on Pro, Max, Team, and Enterprise plans through June 22 — though that access is currently suspended pending the export control resolution.

Claude Code 2.1.170 fixed Fable 5 model names showing a stray [1m] suffix — since Fable 5 ships with 1M context by default, the suffix is now stripped automatically. Also fixed a spurious 'sandbox dependencies missing' startup warning on Windows when sandbox mode was enabled. A separate fix addressed sessions not saving transcripts (and not appearing in --resume) when launched from the VS Code integrated terminal or shells that inherited Claude Code environment variables.

Tata Consultancy Services became a Global Premier Partner in the Claude Partner Network, with Claude rolled out to 50,000 TCS employees across 56 countries in engineering, finance, legal, marketing, and sales. TCS will act as 'customer zero,' refining Claude-powered products before offering them to clients in financial services, healthcare, public sector, aviation, telecom, and medtech. Training and certification will run through TCS iON, which delivers over 75 million assessments annually. Early deployments include Diligenta (TCS's UK life and pensions business) using Claude for 22 million policyholders.

DXC Technology announced it will integrate Claude into the core systems used by banks, airlines, and other regulated industries it serves — joining TCS, PwC, KPMG, Accenture, Cognizant, and Deloitte in Anthropic's rapidly expanding enterprise systems-integrator roster.

Anthropic added roughly 150 new organizations to Project Glasswing — its defensive cybersecurity initiative built on Claude Mythos Preview — bringing the total to about 200 partners across more than 15 countries. The expansion adds coverage in power, water, healthcare, communications, and hardware: critical-infrastructure sectors underrepresented in the initial ~50-partner cohort (which included Apple, Amazon, Broadcom, CrowdStrike, Microsoft, and NVIDIA). That original cohort has already found over 10,000 high- or critical-severity vulnerabilities using Mythos Preview. The expansion was developed in coordination with the US government, the security industry, and open-source maintainers.

In the brief window Fable 5 was available, independent developer Simon Willison published initial impressions and a follow-up noting the model's unusually proactive behavior compared to previous Claude generations — taking more initiative on tasks without being explicitly prompted at each step. Reactions across X were polarized: some called it the new ceiling for AI capability, others focused on the implications of a Mythos-class model reaching general availability at all.