Morning Edition · Friday, June 26, 2026
Geopolitics
US-Led Pax Silica Bloc Widens to 24 States as AI Supply Chains Split Along Geopolitical Lines
European Union members joined the American initiative to govern AI hardware and model supply, weeks after Washington ordered access to two Anthropic frontier models suspended under export controls.
Markets
Agents Become the First Real Economic Test of the AI Buildout
A bank forecast of a roughly 24-fold rise in AI token consumption by 2030, paired with new lab research on agent-driven work, frames inference demand as the measure that will validate or undercut the capital spending.
Tech
IBM Discloses 0.7-Nanometer Transistor Process Using Vertically Stacked Devices
The company describes a three-dimensional nanostack architecture at what it calls 7-angstrom scale, pointing to continued density gains beyond today's leading nodes.
Tech
MindOn Shows a Single Model Controlling Robots of Different Body Types
A demonstration of the Mind-0 system coordinated two humanoid robots and two dual-arm platforms on a shared logistics task, a test of control across different robot bodies (cross-embodiment control).
Tech
Unitree Prices a Humanoid Robot From 4,900 Dollars
The R1, built for research and development, lowers the entry price for a walking humanoid platform and signals the early commoditization of robot hardware.
Tech
New Interpretability Work Locates Sycophancy and Refusal in the Same Activation Geometry
Two papers argue that misbehaviors engineers try to suppress are governed by linear directions in the residual stream, and that persona and refusal are not independent.
Tech
Papers Ask What to Do After a Benchmark Saturates and How to Tell Knowing From Guessing
Two evaluation studies argue that retiring saturated benchmarks discards information and that current tests confuse supported answers with lucky guesses.
Crypto
A New Benchmark Tests Whether AI Agents Can Actually Secure Smart Contracts
CyberChainBench evaluates language-model agents on detecting, exploiting, and patching real on-chain vulnerabilities drawn from hundreds of past exploits.
Tech
Study Finds Standard Post-Training Can Erode Values Instilled During Pretraining
Optimizing models to be helpful through supervised fine-tuning and reinforcement learning degraded compassion-related behaviors in some domains, the paper reports.
Geopolitics
Anthropic Pushes Policy Proposals Built Around an AI Capability Curve It Calls Exponential
The lab argues that policymaking institutions were built for a slower world and proposes changes to match the speed of the technology, the latest sign of frontier labs building their own policy apparatus.
World
Survey Data Show Gen Z Both the Heaviest AI Users and the Most Skeptical
Reported figures put usage well above the share that believes the technology is good for society, a divergence between adoption and trust.
Tech
Researchers Propose Turning Editing Models' Own Moderation Filters Into a Defense
MIRAGE adds perturbations to images so that AI editing systems refuse to alter them, an inversion of standard image-immunization approaches.