America's First AI Software Export Ban Leaves Anthropic's Flagship Models Dark
What Happened: A Two-Step Shutdown
The US export control directive against Anthropic's Fable 5 and Mythos 5 was not a single decision but the result of two compounding security incidents. First, the White House identified SK Telecom, South Korea's largest carrier and a $100 million Anthropic investor, as a risk: the carrier had early access to Mythos 5 capabilities through Project Glasswing, Anthropic's classified cybersecurity consortium, and its parent SK Group held historical business ties to China through a stake in China Unicom. The administration ordered Anthropic to revoke SK Telecom's access. Then Amazon security researchers discovered a jailbreak technique allowing Fable 5 to behave as the unrestricted Mythos 5, potentially exposing powerful automated vulnerability-discovery tools. Amazon CEO Andy Jassy alerted the White House directly. White House AI adviser David Sacks gave Anthropic CEO Dario Amodei an ultimatum: patch the issue or de-deploy. Amodei refused both, calling the bypass narrow rather than a full jailbreak. On June 12, Commerce Secretary Howard Lutnick invoked the Export Controls Reform Act of 2018, barring all foreign nationals globally from accessing either model. Anthropic chose to disable both for every user worldwide rather than attempt selective filtering, and as of June 19, day seven of the shutdown, negotiations with the Trump administration are ongoing.
Why It Matters: Code Is Now a Controlled Export
This is the first time the US government has extended export controls beyond hardware to an AI model, and the implications are as large as they appear. Until now, the export control playbook focused on chips: restrict NVIDIA A100s and H100s, throttle TSMC production capacity, control the physical substrate. Fable 5 changes that playbook permanently. Any frontier AI model can now be treated as controlled technology subject to Commerce Department authority, creating a category of software that may require government clearance before foreign nationals can access it, regardless of geography. The Amazon angle compounds the strangeness: Anthropic's single largest investor and primary cloud infrastructure provider triggered the shutdown of its two most advanced products. That tension between commercial obligation and national security interest will not resolve quietly. For every enterprise that built production workflows on Fable 5 or Mythos 5, the past week is a live stress test of sovereign AI risk: one directive and the model is gone, globally, immediately. The Fable 5 episode also reveals the fragility of the voluntary pre-deployment framework established in the June 2 Trump Executive Order on AI Innovation and Security. Voluntary frameworks do not stay voluntary after incidents like this.
What to Watch Next
Three things will determine how consequential this episode becomes. First, the restoration conditions: Anthropic's international managing director said at the June 18 Seoul office launch that the models would return "in coming days," but what Anthropic concedes to the Commerce Department, whether a government pre-review window, a formal jailbreak audit, or restricted partner tiers, will set the template for every frontier lab going forward. Second, the precedent for other labs: OpenAI, Google DeepMind, and xAI all operate internationally with foreign national employees and users. If Lutnick's June 12 directive becomes the new normal, the legal and operational overhead of deploying a frontier model outside a US-only user base grows significantly. Third, the IPO timing: Anthropic filed its S-1 confidentially on June 1, targeting a $900 billion valuation on projected annualized revenue of $26 billion. An extended shutdown of its two flagship models, combined with a public dispute with the White House, is precisely the regulatory overhang that reprices a deal. Watch whether underwriters hold the valuation or quietly revise the roadshow.
Also worth knowing
- Grok 4.3 reaches GA on Amazon Bedrock, lands natively on Databricks Agent Bricks: xAI's Grok 4.3, with 1M-token context and configurable reasoning, is now available via AWS Bedrock and Databricks' enterprise agent platform, announced at the Databricks Data + AI Summit on June 18. [link]
- Anthropic opens Seoul office, signs AI safety MOU with Korea's Ministry of Science and ICT: Anthropic's third Asia-Pacific office launched June 18 with a government MOU on AI safety and cybersecurity, plus enterprise deals with NAVER, Samsung SDS, LG CNS, and Nexon, a notable charm offensive conducted while the company's top models remain offline. [link]
- Anthropic confidentially files S-1 IPO prospectus with the SEC: Anthropic submitted a confidential S-1 to the SEC on June 1, targeting a $900 billion valuation and projecting $26 billion in annualized revenue, making the Fable 5 export ban particularly damaging in timing. [link]
