# Export Controls Hit Anthropic as AI Scrutiny Broadens

*By AI News Digest • June 13, 2026*

A U.S. export-control order forced Anthropic to disable Fable 5 and Mythos 5 for all customers, prompting a broader debate over talent, access, and power concentration. New York also subpoenaed OpenAI, while NVIDIA and Google DeepMind advanced agentic and robotics infrastructure.

## The story driving today’s cycle

### U.S. export controls abruptly take Anthropic’s newest models offline

The U.S. government, citing national security authorities, ordered Anthropic to suspend access to **Fable 5** and **Mythos 5** for any foreign national, including Anthropic employees, whether they are inside or outside the United States [^1]. Anthropic said the practical result was an immediate shutdown of both models for all customers, while access to other Claude models remains unaffected [^1].

> "We believe this is a misunderstanding and are working to restore access as soon as possible." [^1]

The order lands just days after Fable 5 launched on June 9 as a new tier above Opus, and after Anthropic had already reversed part of its rollout by making certain model redirects visible following backlash [^2].

*Why it matters:* The directive affects both customer access and which employees can use the models, bringing export-control policy directly into frontier-model operations [^1].

### The reaction quickly expanded beyond Anthropic itself

The strongest reactions focused on workforce structure and power concentration. Nathan Lambert said that, in his experience, a minority of LLM researchers are American citizens and warned that rebuilding frontier AI research around citizenship segregation would be "industry destroying" [^3]. Jeremy Howard said he opposed the government action but argued Anthropic should have expected a response after advancing a "too dangerous for anyone except us" posture [^4], while Hugging Face CEO Clement Delangue said he is going to Washington next week to argue for open-source AI, transparency, and against concentration of power [^5].

*Why it matters:* The response tied this single order to broader questions about international staffing, open access, and who gets to shape AI policy [^3][^4][^5].

## Scrutiny is widening to AI product behavior

### New York subpoenas OpenAI on data, engagement, and model conduct

New York’s attorney general has issued a broad subpoena to OpenAI seeking documents related to advertising, user engagement and retention, handling of consumer and health data, activities involving minors and seniors, deep learning models, model sycophancy, and company policies [^6][^7].

*Why it matters:* This inquiry reaches beyond abstract model safety to concrete questions about product design, data handling, vulnerable users, and behavioral effects [^7].

## The buildout around agents and robotics kept moving

### NVIDIA’s Blackwell leads the first AgentPerf benchmark

Artificial Analysis’s new **AgentPerf** benchmark measures how many coding-agent tasks a system can support simultaneously while meeting responsiveness and output-rate thresholds, using long-horizon trajectories drawn from public code repositories across 12+ programming languages [^8]. NVIDIA said its **GB300 NVL72** led the first published results on DeepSeek V4 Pro workloads and ran up to **20x more agents per megawatt** than HGX H200, with Together AI, DeepInfra, and Baseten already serving production agentic workloads on Blackwell [^8].

*Why it matters:* The benchmark compares concurrent agent tasks under latency and token-rate constraints, not just single-turn generation [^8].

### Google DeepMind launches a European robotics accelerator

Google DeepMind said its new **Robotics Accelerator** has launched with 15 startups working on physical AI in Europe [^9]. The three-month program gives participants access to DeepMind’s AI stack, Gemini Robotics models, and hands-on support from its teams [^9].

*Why it matters:* DeepMind is pairing robotics models with direct startup support, extending its AI stack into company-building around physical AI [^9].

---

### Sources

[^1]: [𝕏 post by @AnthropicAI](https://x.com/AnthropicAI/status/2065597531644743999)
[^2]: [AI News: An INSANE Week… Here’s What Matters](https://www.youtube.com/watch?v=nydHKXjwu0U)
[^3]: [𝕏 post by @natolambert](https://x.com/natolambert/status/2065616536942088581)
[^4]: [𝕏 post by @jeremyphoward](https://x.com/jeremyphoward/status/2065636302205452654)
[^5]: [𝕏 post by @ClementDelangue](https://x.com/ClementDelangue/status/2065626715561263290)
[^6]: [𝕏 post by @keachhagey](https://x.com/keachhagey/status/2065580158338113898)
[^7]: [𝕏 post by @GaryMarcus](https://x.com/GaryMarcus/status/2065667991258505559)
[^8]: [NVIDIA Blackwell Leads on First Agentic AI Infrastructure Benchmark](https://blogs.nvidia.com/blog/nvidia-blackwell-agentperf-artificial-analysis)
[^9]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2065388989146628563)