US Bans Foreign Access to Anthropic's Fable 5; China Commits $295B to Domestic AI Grid

The U.S. government halts foreign access to Anthropic's most capable AI models as China commits $295 billion to build a domestic AI computing grid without American chips.

🔐 US Orders Anthropic to Disable Fable 5 and Mythos 5 for All Foreign Users

Decoded: Anthropic announced on June 13 that it will "abruptly disable" its Fable 5 and Mythos 5 AI models for all foreign nationals after receiving a U.S. government export control directive citing national security concerns. The government told Anthropic it believes a method exists to "jailbreak" Fable 5 in a way that would allow it to identify software vulnerabilities — a capability officials say could enable sophisticated cyberattacks on critical infrastructure, including legacy banking systems. Anthropic publicly disputed the justification, saying the government provided only "verbal evidence of a potential narrow, non-universal jailbreak" and that the order did not follow "principles of fair and fact-based regulation." Fable 5 was launched earlier this week as the flagship model in a new capability tier Anthropic calls Mythos-class. The directive arrives as a separate White House dispute over Anthropic's refusal to enable autonomous weapons use had shown signs of resolution ahead of the company's IPO. The Pentagon's chief information officer publicly backed the order. (Reuters, CNA, Nikkei Asia, June 13, 2026)

Why it matters: This is the first time the U.S. government has directly ordered an AI lab to restrict access to a deployed commercial model on national security grounds — extending export control logic from chips to AI software itself. For investors tracking the AI sector, the action creates two immediate risks: it damages Anthropic's international revenue base at a critical pre-IPO window, and it establishes a precedent that Washington can override commercial deployment decisions at any AI lab if it identifies a security concern, however contested. For Google (GOOGL), which holds a major Anthropic stake, the episode introduces headline risk ahead of the IPO roadshow. Longer-term, the jailbreak-as-export-control framing signals a new regulatory lever that competitors — including OpenAI and Google DeepMind — will need to factor into frontier model deployment planning globally.

🏛️ China Commits $295 Billion to Build National AI Computing Grid by 2028

Decoded: China's National Development and Reform Commission is mobilizing approximately 2 trillion yuan ($295 billion) over five years to build a unified nationwide AI data center network targeting completion by 2028, according to details reported June 12. The plan mandates that at least 80% of hardware and software powering the network — including AI chips — come from domestic Chinese suppliers, with Huawei positioned as the primary beneficiary. State-owned telecom operators China Mobile and China Telecom will lead construction and interconnection. The initiative expands China's existing "Eastern Data, Western Computing" infrastructure framework launched in 2022 and will be financed through a mix of sovereign debt, state funds, bank loans, and private investment. (Reuters, June 12, 2026; NDRC official documentation)

Why it matters: China's $295 billion commitment is a direct structural response to U.S. semiconductor export controls — the 80% domestic sourcing mandate effectively treats Huawei's Ascend AI chips as the reference platform for the entire national computing infrastructure, bypassing Nvidia (NVDA) and AMD hardware entirely. For investors in Nvidia, the scale of this locked-out market is now quantifiable: $295 billion in AI compute investment over five years, none of it accessible to U.S. chip suppliers. The deeper implication is technological bifurcation: a fully sovereign Chinese AI compute grid, if it reaches the planned 2028 completion, would give Beijing AI inference capacity that operates entirely outside U.S. technology supply chains and export control jurisdiction — a structural shift in the global AI competitive landscape that goes well beyond near-term revenue loss for American semiconductor companies.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team