The Trump administration's sudden decision to bar non-US persons from accessing Anthropic's most advanced artificial intelligence models has sent ripples across the Pacific, forcing the company to open a new office in Seoul while major American investment banks preemptively cut off overseas staff from its tools. The restrictions on the Fable 5 and Mythos 5 models, reported by The Washington Post, were triggered after US officials reviewed an expanded access list of roughly 150 institutions and identified a South Korean telecoms company they suspected of maintaining links to China. Though unnamed in official accounts, market attention has focused on SK Telecom, which had recently secured early access to Mythos through Anthropic's Project Glasswing. The firm has denied any connection to Chinese entities, stating it does not use Huawei equipment in its core network, but the damage to trust was already done.

Viewed from Washington, the episode reflects a hardening resolve to treat frontier AI systems as strategic assets requiring strict export controls. The White House has reportedly issued an ultimatum to Anthropic: the curbs will only be lifted once the company patches jailbreaking vulnerabilities that allow users to bypass safety guardrails through cleverly crafted text prompts. According to Wired, officials made clear during a high-level technical meeting on 15 June that they were no longer interested in debating the real-world severity of the risks; Anthropic's earlier insistence that the concerns were overblown had exhausted official patience. The administration's stance signals that access to cutting-edge AI will increasingly be conditioned on demonstrable security hardening, not merely corporate assurances.

In Asia, the fallout has been swift and extends well beyond the telecoms sector. JPMorgan Chase and Goldman Sachs have each instructed employees outside the United States to stop using Anthropic's Claude models. Staff at JPMorgan's Hong Kong office can no longer select Claude from an internal dropdown list of approved large language models, a decision the Financial Times reports was based on the wording of Anthropic's licensing terms. Goldman Sachs had earlier adopted an equally strict interpretation, effectively banning usage in Hong Kong to avoid any risk of the technology reaching mainland China. These moves, while not directly ordered by Washington, illustrate how US financial institutions are proactively aligning their internal compliance with the spirit of export controls, treating any non-US access as a potential vector for leakage.

Anthropic's response has been twofold: publicly, it is deepening its footprint in a market its managing director for international operations described as "one of our fastest-growing in the world," citing Korea's technical base and developer talent. The new Seoul office is intended to serve as a hub for regional clients and a signal of commitment. Privately, the company is under pressure to demonstrate that its software can withstand adversarial probing. The intersection of these dynamics—geopolitical suspicion, corporate compliance, and technical vulnerability—is reshaping the landscape for AI deployment. Analysts in London note that the episode may accelerate a balkanisation of AI services, with providers forced to create jurisdiction-specific instances and audit trails, while Asian technology hubs risk being caught between American innovation and Chinese-linked supply chains. The long-term question is whether such controls will safeguard sensitive capabilities or simply spur the development of alternative ecosystems beyond Washington's reach.