One week after the White House forced Anthropic to pull its most powerful AI models offline for every user on the planet, the company opened a new office in Seoul and told a room full of Korean government officials and tech executives that the models would be back soon. The juxtaposition captures the strange position Anthropic now occupies: simultaneously the hottest AI company in the world and one under active suspension by its own government.
Seoul as a Strategic Signal
Anthropic launched its Seoul office on Wednesday, led by KiYoung Choi, a three-decade veteran of the Korean tech industry. The company signed a memorandum of understanding with South Korea’s Ministry of Science and ICT to support AI adoption across the Korean public sector.
The timing was not accidental. The Seoul expansion comes days after the export crisis was triggered in part by SK Telecom, one of Korea’s largest telecommunications companies. White House officials identified SK Telecom as having suspected ties to China and ordered Anthropic to revoke the company’s access to Claude Mythos, which Anthropic did immediately. The subsequent discovery of security concerns in Fable 5 compounded the situation, and the Trump administration ordered both models taken offline entirely on June 12.
Opening a Seoul office in that context is a deliberate move to separate the company’s relationship with the broader Korean AI ecosystem from the SK Telecom incident. Anthropic is effectively telling Seoul: the problem was one company with alleged China connections, not Korea itself.
The “Coming Days” Restoration Timeline
At the Seoul launch event, Anthropic’s international managing director told reporters he was “very confident” that both Fable 5 and Mythos 5 would return “in the coming days.” Negotiations between Anthropic and the Trump administration have been running continuously since the June 12 shutdown, and the company’s public posture suggests those talks are going well.
What Anthropic has not said is whether the restored models will come back with the same access terms. The original Fable 5 and Mythos 5 were available to roughly 150 organizations through Project Glasswing, Anthropic’s restricted advanced-access program. If the restored version ships with tighter geographic restrictions or new compliance requirements for participants, that changes the commercial calculus for every enterprise customer in the pipeline.
The Business Impact Is Already Real
The shutdown has forced Anthropic’s enterprise customers into contingency mode. Companies that had built workflows around Claude Mythos or Fable 5 have had to fall back to older Claude models or switch to competitors entirely. When BTN reported on the initial shutdown, the immediate concern was developer trust. A week later, the concern has shifted to contract terms: enterprise buyers are now asking what happens if a government can pull a model with 24 hours’ notice.
That question matters more than it might seem. Anthropic’s annualized revenue has been growing at extraordinary rates, crossing $47 billion in run-rate by mid-May. Much of that revenue comes from enterprise API contracts where reliability and uptime guarantees are fundamental. A precedent that the U.S. government can force a model offline for all users, including domestic ones, introduces a risk factor that no enterprise procurement team can ignore.
The Geopolitics of AI Model Access
The Fable 5 episode is the first major test of how AI export controls work in practice. Previous restrictions on AI chips and training compute were about hardware, physical goods that can be tracked at borders. Model access controls are fundamentally different: they require the AI company itself to act as an enforcement mechanism for government policy.
Anthropic complied instantly when asked to revoke SK Telecom’s access. It complied again when told to take both models offline entirely. That compliance speed may have preserved the company’s relationship with Washington, but it also demonstrated to every international customer that Anthropic can and will cut access on government orders, with no advance warning and no appeal process.
For the broader AI industry, the implications run deeper than one company’s bad week. OpenAI, Google, and every other frontier lab now knows that their most advanced models can become tools of trade policy at a moment’s notice. The question for 2026 is whether that power will be used surgically or as a broad instrument, and whether international customers will start demanding contractual protections against government-ordered shutdowns.
The Seoul office opening suggests Anthropic is betting on the surgical approach. If the models come back with targeted restrictions rather than blanket bans, the export control framework may actually work as intended: protecting genuine security interests without destroying the commercial relationships that fund the research. If they don’t, the Seoul office will be a very expensive gesture.
The next 72 hours will tell the story. Anthropic’s “coming days” language is not vague optimism from a company with no leverage. It is a company with $47 billion in run-rate revenue telling the White House that every day the models stay offline costs real money, real contracts, and real credibility in exactly the international markets Washington claims to want the U.S. AI industry to dominate.