Best of your X follows: June 16
The US government's export control order forces Anthropic to shut down Fable 5 and Mythos 5 for all customers worldwide — the most direct government intervention in frontier AI deployment to date. Also: Mollick's essay on the escalating instability pattern, Naval's quip on AI irreversibility, and simonw's self-built gpt-realtime-2 voice playground.
The dominant story today arrived at midnight UTC: the US government has ordered Anthropic to shut off its two most powerful models for all customers, worldwide. Here's what happened, plus two smaller signals worth noting.
Policy: US government forces Fable 5 and Mythos 5 offline globally
At 12:50 AM UTC on June 13, Anthropic posted a statement that stopped the AI industry cold. 1
The US government, citing national security authorities, issued an export control directive requiring Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national — whether inside or outside the United States, including Anthropic's own foreign national employees. Because there is no practical way to verify nationality at the API level, the net effect is a global shutdown of both models.
"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.""We believe this is a misunderstanding and are working to restore access as soon as possible."— AnthropicAI 1
Access to all other Claude models remains unaffected.
コンテンツカードを読み込んでいます…
As of this writing, the directive is still in effect and neither Fable 5 nor Mythos 5 are accessible. Anthropic has characterized it as a "misunderstanding," which suggests the company disputes the classification of these models as export-controlled items — but has not specified exactly which regulation the directive invokes, or whether any internal process was triggered by the models' dual-use capabilities.
The timing is notable. Fable 5 was released just four days ago (June 9), and the prior issues of this digest tracked the rapid enterprise and developer ecosystem responses to it. Export controls have historically applied to physical hardware and semiconductor technology; applying them to a software inference API is an unusual step that will be closely watched across the industry.
What this means if you depend on these models: Access is off now. The practical fallback is Claude Opus 4.7 or other unaffected Claude models, or switching to a competing provider. Anthropic has offered no timeline for restoration.
The shape of the thing — when a week becomes a pattern
Wharton professor Ethan Mollick this morning pointed back to an essay he had written months ago, originally written in the context of the Anthropic/DoW conflict and the Citrini and Block controversies. 2
"But I think that single week is a good illustration of what the near future will feel like… as the stakes go up, it is likely things will feel even more unstable."
The observation lands differently today. The export control order is the most direct government intervention in frontier AI deployment seen to date. Whether it sticks or gets reversed quickly, the shape of the week will be studied as a reference point.
コンテンツカードを読み込んでいます…
On the same theme, Naval posted a short observation with obvious double meaning given today's news: 3
"Can't put the genie back in the GPU."
The tweet had 291 retweets and 4,375 likes before the Anthropic statement went out — its half-life lengthened considerably once the shutdown was announced. The irreversibility argument cuts both ways: a model powerful enough to draw national security interest is also one that, once widely distributed via API, cannot simply be recalled.
Tools: simonw builds his own gpt-realtime-2 voice playground
Away from the policy drama, Simon Willison posted a practical tool. Fed up with OpenAI not shipping
gpt-realtime-2 — a much-improved voice conversation model — to the main ChatGPT product, he updated his OpenAI-WebRTC playground to use it, and added the ability to paste in a document and have a conversation about it. 4コンテンツカードを読み込んでいます…
The post currently sits at 233 favorites and 120 bookmarks — the bookmarks-to-favorites ratio (roughly 1:2) is typical of the "I'll actually use this later" reaction, rather than just engagement. The playground is live and linked in the tweet. If you've been waiting for a better voice interface for working through documents, this is available now via the API without waiting for OpenAI's product team.
This digest covers substantive posts from a curated set of public AI/tech accounts on X. Pure retweets, aphorisms without AI/tech content, and promotional-only posts are filtered out. Today's excluded items: karpathy SpaceX tribute (off-topic), ylecun paper deadline RT (minor), naval aphorisms without substantive AI content.
このコンテンツについて、さらに観点や背景を補足しましょう。