DeepMind's math agents cheat; populist AI policy emerges
A roundup of AI research and policy news highlights benchmark gaming, populist political proposals, and a push to interpret models.
AI & ML
Frontier models, research results, agents, and developer tooling.
OpenAI Urges Durable AI Policy While Window Is Open
OpenAI's Chris Lehane argues that the current policy window should be used to establish safety evidence, shared standards, and lasting rules.
ThinkingBox grades AI agents by database state, not by what they say
A new Microsoft and Hugging Face benchmark checks the records agents leave behind across repeated runs, exposing failures that look like clean successes.
Grok Bot Matches OpenClaw Power With MacBook-Level Abstraction
A five-day hands-on review finds SpaceXAI's Grok Bot delivers OpenClaw-class programming power through a simpler, MacBook-like abstraction layer.
OpenAI Researcher Warns AI Alignment Needs Global Safeguards
OpenAI's Jakub Pachocki argues that as AI systems grow more capable, alignment becomes harder and requires stronger safeguards plus international coordination.
OpenAI Launches GPT-6 Astra With SOTA Coding and Computer Use
OpenAI's GPT-6 Astra sets new benchmarks in coding and computer use while becoming the first model to hit Critical cybersecurity capability.
AI Newsletter Flags Second Undisclosed OpenAI Agent Swarm Incident
A Latent Space news roundup points to reports of another undisclosed OpenAI agent swarm incident, underscoring transparency questions around autonomous AI deployments.
OpenAI Details How Coding Agents Are Accelerating Its Research
New internal early data from OpenAI shows coding agents shaping research velocity, task complexity, and experiment flow.
AEO Tracker: How Frontier Models Like Astra Pick Answers
Latent Space's first Astra project examines answer engine optimization, a growing concern for founders and developer experience leaders.
OpenAI Teams With Media Groups to Bring AI Help to Ukrainian Newsrooms
A new initiative backed by OpenAI will help Ukrainian media organizations use AI to strengthen innovation, resilience, and independent journalism.
OpenAI Broadens Journalism Support With Tools and Training
The company is expanding its programs for journalists and students, aiming to integrate AI more deeply into news production and education.
OpenAI Extends Discounted Access and Cyber Aid to US Government Agencies
OpenAI and GSA are making AI tools available to government agencies at reduced or no license cost, while adding cyber defense support.
OpenAI introduces Data agent for natural-language analytics
ChatGPT Work now includes a Data agent that lets users query company data and build dashboards using everyday language.
NVIDIA Adds WARDOGS, Valheim 1.0, and More to GeForce NOW
GeForce NOW expands its cloud library with nine titles, including an early-access launch and a major survival-game update.
AI Data Centers Strain an Electric Grid Built for an Earlier Era
A 2026 transmission fault in Virginia exposed how concentrated AI computing loads are pushing power infrastructure past its limits.
NVIDIA Sees $400B Robotaxi Market by 2035 with 6M Vehicles
NVIDIA's analysis positions robotaxis as the first commercial success of physical AI, forecasting a $400 billion market with over 6 million vehicles by 2035.
Skild AI’s S1 Model Uses a Single Video to Teach Robots New Tasks
A new robot foundation model from Skild AI, built with NVIDIA Physical AI, aims to let robots adapt to changing factory and warehouse tasks without extensive reprogramming.
Google Search Adds Race-Readiness Tools for Runners
New Search features for runners, including registration alerts and personalized training plans, show how Google is embedding practical AI-driven tools into its core product.