White House Accelerates Migration to Post-Quantum Cryptography

A US executive order, "Securing the Nation against Advanced Cryptographic Attacks," has dramatically shortened the deadlines for migrating the cryptographic systems of government agencies and critical infrastructure to schemes that can withstand quantum computer threats. Key establishment methods for "high-value assets" must be migrated to post-quantum cryptography by December 31, 2030, and digital signature schemes by December 31, 2031.

This amounts to pulling the schedule forward by as much as five years from the previous timeline. Behind the move is research showing that the cost and resources required to build a cryptographically relevant quantum computer are significantly lower than earlier estimates suggested. Google and Cloudflare have already moved their migration targets up to 2029, and the industry as a whole is accelerating its response.

With military, financial, government, and personal confidential data all potential targets of "Store Now, Decrypt Later" attacks, updating cryptographic infrastructure has become an urgent challenge that can no longer wait.

Linux Foundation to Build a Trusted Identity Foundation for AI Agents with the "Agent Name Service"

The Linux Foundation has announced the launch of "**Agent Name Service (ANS)**," a trusted identity foundation for AI agents. It aims to serve as the infrastructure through which AI agents are securely identified and authenticated on the internet, playing the same role for AI agents that DNS plays in providing domain names for websites.

As agentic AI becomes widespread, a mechanism for verifying "who operates this AI agent and what it can do" is essential, yet no standardized framework exists today. ANS is drawing attention as an attempt to fill this gap.

Detailed specifications and launch timing remain at the announcement stage for now; further developments are eagerly awaited.

SoftBank's ¥122 Billion AI Supercomputer — and Why It Still Chose NVIDIA GPUs

Developers have shared the behind-the-scenes story of SoftBank's ¥122 billion AI supercomputer, revealing a trial-and-error process so intense that they recall being "freaked out at first by how much it broke."

As for why NVIDIA GPUs were nevertheless chosen, the developers point to the maturity of the CUDA ecosystem and its overwhelming optimization for AI workloads. In building their own AI infrastructure, it wasn't just raw hardware performance — the stability of the entire software stack and development efficiency became the decisive factors.

China's Humanoid Robot Market: Shipments Grow Roughly 7x Year-over-Year

According to an analysis by Li Zhihui of Nomura Research Institute, China's humanoid robot market has entered the mass-production phase. Shipments have grown roughly sevenfold year-over-year, and the country now accounts for about 80% of the global market share.

The number of robot manufacturers has doubled as companies from other industries move in, with automakers, home appliance makers, and major tech companies all rushing into the humanoid field. Combined with Chinese government policy support and a concentrated supply chain, China is establishing itself as the world's production hub for humanoid robots.

"No AI Co-Authors" Manifesto Stirs Controversy — A Move to Turn Away from AI Collaboration

"[No AI Co-Authors. A Manifesto](https://no-ai-coauthors.dev)", which gathered 10 points on Hacker News, is stirring debate. The manifesto, which opposes crediting AI as a co-author in writing, calls into question the role of AI in creative work.

Around the same time, an HN thread on "how to find companies that use AI modestly" and a first-hand account of "working with AI for four years as a content writer" have also drawn attention. As business use of AI becomes commonplace, backlash against "leaving everything to AI" is growing among some engineers and creators.

Notably, the prominent voices are not outright AI denialism, but rather a stance of "use AI appropriately, but don't entrust everything to it." While initiatives like CyberAgent's drive to permeate AI throughout the organization (including a contest with a ¥10 million prize and an "AI Banzuke" ranking) move forward, the struggle over how to use AI is becoming a reality as well.

650+ Open-Source Biomedical NER Models Run at High Speed on MLX

The OpenMed project has released more than 650 biomedical named entity recognition (NER) models under the Apache 2.0 license. They run on Apple Silicon's MLX framework, reportedly delivering a 30–40x speedup compared to PyTorch CPU.

On a three-year-old MacBook Pro M3 Max, a 434M-parameter model can perform inference in 27ms (MLX). With the same weights and the same fp32 precision, the output is said to exactly match that of PyTorch CPU at roughly 1080ms. The ability to process personal medical data on-device is also drawing attention from a privacy protection perspective.

---