This Week in AI: June 15–19, 2026.
A roundup of the developments that shaped the week in artificial intelligence — from a precedent-setting U.S. export action to Anthropic's push into Korea, fresh research from OpenAI, and a flagship model still waiting in the wings.
Export controls keep two frontier models offline
The week's dominant story was the continuing fallout from a U.S. Commerce Department export-control directive issued June 12 that barred foreign-national access to Anthropic's two newest models, Claude Fable 5 and Claude Mythos 5. Unable to screen users by nationality, Anthropic disabled both models entirely, calling the action a misunderstanding and saying it was working to restore access; its other Claude models were unaffected.
The models stayed offline through the week, and policy analysts flagged the precedent the move sets for how Washington can compel a lab to pull a deployed model. Competitors moved to court displaced customers — Chinese lab MiniMax promoted the open-weights availability of its M3 model as an alternative that can't be recalled by government order.
Anthropic opens a Seoul office amid the fallout
On June 17, Anthropic formally opened a Seoul office and announced a wave of Korean enterprise deployments — including Claude Code across NAVER's engineering organization, Claude and Claude Cowork at Samsung SDS for Samsung Electronics, and rollouts at LG CNS, Hanwha Solutions, Nexon, and others. The launch landed just days after the export controls cut Korean institutions off from the company's most advanced models — an awkward backdrop several outlets noted.
OpenAI leans into scientific R&D
OpenAI used the week to spotlight science: on June 17 it introduced LifeSciBench, a life-sciences benchmark, and a method for predicting model behavior before release; on June 18 it detailed a near-autonomous "AI chemist" that improved a difficult medicinal-chemistry reaction. It also rolled out a new OpenAI Partner Network.
Google's Gemini 3.5 Pro is still in the wings
Google's most powerful model of the year, Gemini 3.5 Pro — announced at I/O in May with a June target, a 2-million-token context window, and a "Deep Think" reasoning mode — still hadn't reached general availability, remaining in limited enterprise preview as the month's window narrowed.
The AI IPO race stays in the spotlight
Anthropic's confidential SEC filing — submitted June 1 at a roughly $965 billion valuation, with an October listing targeted — drew continued coverage, while rival OpenAI is reported to be preparing its own offering, setting up one of the most closely watched IPO seasons in tech.
What to watch
The bulk of the EU AI Act's obligations take effect on August 2, 2026, including the rules for high-risk AI systems — a deadline that will reshape how companies build and deploy AI in Europe. Together with this week's U.S. export action, it's a reminder of how quickly AI governance is tightening on both sides of the Atlantic. We build on this shifting ground every day — if you're weighing what it means for your own roadmap, we'd love to compare notes.