The Trump administration ordered Anthropic to block foreign nationals from accessing Fable 5 and Mythos 5, forcing the company to take models offline globally. The action reveals complexities in applying export controls to frontier AI models.

Anthropic cannot distribute Claude Mythos and Fable 5 after running afoul of Trump administration export controls, but without clarity on specific violations. The situation illustrates regulatory uncertainty facing AI companies under new export policies.

A Decoder episode examines the regulatory controversy around Anthropic and its advanced Fable 5 models under the Trump administration. The discussion raises fundamental questions about governance, safety, and control in frontier AI models.

The White House ordered Anthropic to revoke SK Telecom's access to Claude Mythos citing alleged China ties, days before models went offline. The incident illustrates how geopolitics impacts access to advanced AI technology.

Trump administration officials demand Anthropic ensure Fable 5's guardrails cannot be circumvented as condition to relaunch the model. Security experts warn completely eliminating jailbreaks may be technically impossible.

Inference startup Baseten is closing a $1.5 billion round at a $13 billion valuation, reinforcing the massive investment trend in AI infrastructure. This move underscores growing demand for optimized inference solutions.

OpenAI hired Noam Shazeer (Transformers co-inventor) from Google DeepMind and Dean Ball, former Trump AI policy official, in the same week. The personnel moves signal significant preparations ahead of OpenAI's planned IPO.

AWS is negotiating to sell its AI chips to other data centers, representing a $50 billion opportunity according to CEO Andy Jassy. The move positions Amazon as a direct Nvidia competitor in the AI hardware market.

Startup General Intuition, specialized in AI agents with spatial-temporal reasoning, is in talks to raise $300 million with backing from investors including Jeff Bezos. The funding reflects growing interest in advanced AI agents with sophisticated reasoning capabilities.

The Federal Energy Regulatory Commission issued a major decision simplifying the connection of AI data centers to the electrical grid. The move is crucial for AI infrastructure expansion, enabling new facilities to connect more easily.

Snap spun off its AI video team to form Dotmo, a new company focused on AI-powered video development. The move reflects how companies are restructuring to manage high AI research costs.

Three Amazon engineers face potential termination after testifying about data center limits to Seattle City Council, accusing retaliation. The case touches tensions between environmental policy, labor activism, and AI infrastructure investment.

NVIDIA showcases at Cannes Lions how its technologies enable autonomous operations in advertising and marketing. The industry is shifting from AI adoption to infrastructure supporting fully autonomous agents.

Meta employees in the newly formed AI unit report significant dysfunction and low morale, according to recent reports. Internal issues at Meta's restructuring reflect integration challenges in AI teams.

France continues deploying its national AI strategy using NVIDIA infrastructure, including AI factories, local compute capacity, and open models. The French AI ecosystem is running production agents and launching applications.

Researchers used OpenAI's reasoning model to identify 18 new diagnoses in previously unsolved cases of rare genetic diseases in children. This breakthrough demonstrates AI's potential in precision medicine and could transform rare disease diagnosis.

OpenAI and Molecule.one demonstrated how an autonomous AI chemist using GPT optimized a key drug-synthesis reaction. This achievement accelerates medicinal chemistry research and opens new possibilities for drug discovery.

Research published in Nature shows Google's AMIE conversational medical AI system matches primary care physicians in complex disease management cases. This validates the potential of conversational AI in real clinical settings.

Hugging Face published an analysis on MosaicLeaks, a vulnerability compromising confidentiality in AI-powered research agents. The finding is important for developers building autonomous systems with access to sensitive data.

Researchers presented cuTile Rust, a framework bringing memory safety to GPU kernels through compiler verification, achieving performance competitive with vLLM/SGLang. The work is relevant for developers wanting safety guarantees in GPU-generated code.

Hugging Face publishes research examining alternatives to LoRA, the most popular fine-tuning technique today. The analysis seeks methods that overcome LoRA's limitations for more efficient model adaptation.

Adobe launches custom AI assistants in Photoshop, Premiere, Illustrator, InDesign, and Frame.io in public beta. This expansion integrates generative capabilities into the industry's most-used creative tools, transforming workflows for millions of designers.

OpenAI launched new analytics and spend controls for ChatGPT Enterprise, enabling organizations to manage costs and scale AI with better visibility. This update is relevant for DevOps teams and technical leaders who need to monitor API consumption.

OpenAI improved ChatGPT's health responses using GPT-4.5 Instant, featuring stronger reasoning and physician-informed evaluations. The update reflects OpenAI's focus on deploying AI in regulated sectors with greater rigor.

Midjourney announced The Midjourney Scanner, its first hardware product: a full-body ultrasound scanner using a ring of sensors. The move marks the company's expansion into medical AI hardware.

Adobe introduces a reimagined Firefly AI studio with persistent context, reusable assets, and organized workflows in private beta. The upgrade enables consistent work across multiple design projects.

Apple released a patch for a high-severity eavesdropping vulnerability in Beats Studio Buds, disclosed 12 months ago. The bug affects multiple manufacturers and is a reminder about IoT security.

Hugging Face publishes tools to evaluate whether open models have sufficient agentic capabilities for your specific tooling. The benchmark helps developers select appropriate models for AI agent applications.