South Korea made the largest enterprise commitment to Anthropic's Claude AI coding tools in Asia this month, even as a US export control directive cut off access to the company's most advanced models worldwide. Major Korean technology companies including NAVER, Samsung Electronics, LG Group, Nexon, and Hanwha Solutions deployed Claude Code and related tools across tens of thousands of engineers, marking the most concentrated adoption of Anthropic's coding platform anywhere in the region. Yet the same week revealed a more complex story about how national security concerns reshape AI access for enterprises globally.

What triggered the US export ban on Anthropic's top models?

On June 12, the US Commerce Department issued a directive under the Export Controls Reform Act of 2018, citing national security concerns, and Anthropic immediately disabled access to its Mythos and Fable 5 models for all customers worldwide. The ban stemmed from two separate developments. First, the White House learned that SK Telecom, South Korea's largest wireless carrier and an Anthropic investor, had gained access to Mythos 5 through Project Glasswing, Anthropic's invite-only cybersecurity consortium, and US officials suspected undisclosed ties to China. Second, Amazon, Anthropic's largest investor with approximately $13 billion in cumulative stakes, flagged a security vulnerability in Fable 5 to the White House after researchers discovered that the model could be prompted to identify code vulnerabilities, effectively bypassing its safety guardrails.

The distinction between the models matters significantly. Claude Sonnet 4.6 and Opus 4.8, which powered the Korean deployments, remained fully available throughout the ban. The restricted models, Fable 5 and Mythos 5, represented a different class of capability. Mythos is Anthropic's security-focused model, described at its April 2026 launch as the first AI system to complete both cybersecurity test ranges that the UK AI Security Institute uses to evaluate offensive hacking capability.

Why did the ban affect all users, not just foreign nationals?

The global outage resulted from a structural gap in how export control law interacts with cloud-delivered AI. The "deemed export" doctrine under 15 CFR 734.13 treats the release of controlled technology to a foreign national inside the United States as an export to that person's home country. This rule was originally written for semiconductor blueprints and weapons schematics, physical artifacts that can be inventoried and licensed individually. A cloud API endpoint serving hundreds of millions of users has no nationality field in its session tokens or request headers, making real-time verification of user citizenship impossible. Anthropic's only compliance option was to shut off access for everyone.

The impact extended far beyond Korea. Roughly 200 elite firms across 15 countries lost access to Mythos Preview, the unguarded version of the model priced at $25 per million input tokens and $125 per million output tokens. These organizations, which included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, had been using the model to scan their own code for security flaws.

What capabilities did enterprises lose?

Between April and late May, Mythos Preview surfaced more than 10,000 high or critical severity vulnerabilities across the consortium, according to Anthropic. Individual organizations reported significant findings: Cloudflare identified 2,000 bugs across critical systems, 400 of them rated high or critical, while Mozilla patched 271 vulnerabilities in Firefox 150 while testing the model. For security teams cut off by the directive, the loss of access to Mythos represented a meaningful gap in their defensive capabilities. These teams have since shifted to Opus 4.8 as a fallback, a model Anthropic itself does not call its strongest cyber defender.

How did the NSA testimony reshape the debate?

On June 21, reporters clarified claims that had circulated widely about Mythos breaching classified systems. Shashank Joshi, defense editor at The Economist, confirmed on social media that he had accurately quoted Senator Mark Warner, vice chair of the Senate Intelligence Committee, but said the line had been stripped of crucial context as it spread across social media. Warner, speaking on June 11, relayed what General Joshua Rudd, the four-star officer who jointly leads the NSA and US Cyber Command, had told him in a private briefing: Mythos "broke into almost all of our classified systems, not in weeks, but in hours" during authorized testing. The exercise was a red team drill on government networks, not an outside intrusion, and Warner cited it to argue for faster pre-release testing of frontier AI models, not to condemn Anthropic.

Even with those caveats, the disclosure confirmed that the US government's concern extended beyond the narrow "fix this code" jailbreak to the underlying autonomous offensive capability of Mythos-class AI. A model that can find and chain vulnerabilities at that speed, in authorized testing against its own government's infrastructure, represents a qualitatively different artifact from prior generations of dual-use AI that export control frameworks were designed to govern.

What does this mean for Anthropic's enterprise strategy?

The Korean deployments demonstrate Anthropic's strength in enterprise adoption, particularly in coding and knowledge work automation. NAVER, Korea's dominant internet platform and search engine, deployed Claude Code across its entire engineering organization, marking the largest single enterprise adoption of Claude Code in Asia. Samsung SDS, the IT services arm of Samsung Group, introduced both Claude CoWork and Claude Code to Samsung Electronics employees, covering knowledge work, agentic workflows, and large-scale software development. These deployments reflect Anthropic's competitive positioning in high-tech-maturity enterprises where safety architecture and compliance depth create defensible advantages.

However, the export ban exposed vulnerabilities in Anthropic's business model. The company's revenue model is enterprise-first and API-driven, with approximately 80 percent of revenue from business customers. Anthropic had achieved approximately $47 billion in annual recurring revenue as of May 2026, up from $9 billion at the end of 2025, representing roughly 10 times year-on-year growth for three consecutive years. Yet the ban demonstrated how geopolitical restrictions can instantly cut off access to revenue-generating products for hundreds of organizations simultaneously.

How to understand Anthropic's competitive position in enterprise AI?

• Model Architecture: Anthropic offers a tiered model lineup with Opus for high-complexity reasoning, Sonnet as the enterprise production workhorse, and Haiku for high-volume latency-sensitive applications, allowing enterprises to match model capability to specific workload requirements.
• Product Velocity: Claude Code reached $2.5 billion in annual recurring revenue as a standalone product within nine months of launch, representing the fastest B2B product ramp in software history and demonstrating strong market demand for agentic coding tools.
• Distribution Reach: Anthropic is the only frontier model available natively across AWS Bedrock, Google Vertex AI, and Microsoft Azure Foundry simultaneously, removing procurement blockers that typically favor hyperscaler-owned models.
• Enterprise Customer Base: Anthropic serves 1,000 enterprises spending over $1 million annually, with 8 of the Fortune 10 as Claude customers, indicating deep penetration in high-value accounts.
• Safety as Moat: Constitutional AI, Anthropic's approach to embedding safety at the model level, makes Claude the default choice in regulated industries including financial services, healthcare, and legal, where auditability is a procurement prerequisite.

Anthropic is now drafting a shared risk framework with the White House, according to Politico, with the proposed standard designed to score how far a bypass went, what capabilities became reachable, and whether the exposed behavior had real operational effects. President Donald Trump told Axios on June 21 that he no longer viewed Anthropic CEO Dario Amodei as a national security threat after meeting him at the G7 summit, calling him "nice" and "smart". This shift suggests potential movement toward a resolution that could restore access to restricted models for vetted enterprises.

The Korean deployment story illustrates a broader tension in the AI industry: enterprises want access to the most capable models for competitive advantage, but national security frameworks designed for physical weapons are struggling to accommodate cloud-delivered AI systems that serve global customers simultaneously. For Anthropic, the challenge is demonstrating that its safety architecture and compliance depth can satisfy government concerns while maintaining the enterprise relationships that drive its explosive growth.