Anthropic released a comprehensive threat report detailing global misuse of its Claude model, including state-sponsored cyber operations and unauthorized training by Chinese labs, amid high-profile safety researcher resignations over existential risks. Concurrently, OpenAI launched ChatGPT for Financial Services powered by GPT-6 Astra alongside its new Agents API, while Chinese rival DeepSeek released its highly efficient V4.1-Flash open-weight model.
Sources: HuggingNews, Michael Parekh AI, OpenAI Blog, Simon Willison, The Neuron, The Rundown AI
Oracle reported a 121% jump in cloud infrastructure revenue to $7.4 billion, driven by massive AI demand that prompted the company to expand its capacity with over 300,000 GPUs and secure $11 billion in customer prepayments. This surge aligns with a broader industry push, as Microsoft plans to triple its data center capacity to over 38 gigawatts by 2032, and SpaceX secures a major $1.1 billion monthly compute rental deal.
OpenAI CEO Sam Altman has discussed potentially pacing the development of advanced frontier AI models and is consulting Congress on whether a coordinated industry-wide slowdown with competitors like Anthropic would violate antitrust laws. The proposed pause stems from growing safety concerns, even as major AI companies face intense competition and financial pressures.
California Governor Gavin Newsom has signed four new laws establishing the nation's first state-level framework for independent AI safety assessments and auditor standards. In response, OpenAI endorsed the legislation while urging the U.S. Congress to implement mandatory federal safety rules that include exemptions for startups and open-weight models.
Officials in the United Arab Emirates are shifting from a centralized Abu Dhabi data hub to a distributed network of smaller sites for their 5 GW AI project following Iranian missile and drone attacks. Led by G42 with partners including Nvidia, OpenAI, and Oracle, the project's security measures now include potential underground facilities and enhanced air defenses.
OpenAI withdrew from a Caltech math hackathon after prominent mathematicians published an open letter criticizing AI-generated math 'slop'. This highlights growing friction between academic communities and AI developers regarding mathematical rigor.
Leopold Aschenbrenner's hedge fund purchased call options totaling hundreds of millions in premiums targeting AI and semiconductor companies like AMD and CoreWeave. This massive market move signals a strong return to the AI sector following the collapse of his previous fund.
Nvidia has introduced SoL-Pi, a standalone extension for the Pi agent based on auto-research loops designed to reduce token traffic, inference work, and redundant agent turns.
Alibaba's Qwen team released Qwen3.8-2.4T-A95B, marking the first time a Qwen-Max-class model has been made available as open weights. Featuring 2.4 trillion total parameters and a hybrid linear-plus-full-attention architecture, it targets demanding agentic and reasoning workloads.
25 Fields Medallists signed a declaration warning that the push by AI companies to solve mathematical benchmark problems threatens the science of mathematics and human intellectual traditions.
Shopify is moving its mobile apps back to separate Swift and Kotlin codebases, citing that AI coding agents now handle enough implementation and translation work to eliminate the traditional trade-offs of native development.
Calif Research demonstrated WeWorm, claiming it is the first zero-click worm to spread via WeChat calls on iOS and Android. The team stated that AI tools helped them discover the bug and write the remote code execution exploit in just two days.
Iranian forces recovered a malfunctioning Anduril Dive-LD underwater vehicle near the Strait of Hormuz, prompting analysts to suggest Iran may reverse engineer its mechanical systems to improve its own unmanned designs.
A new paper introduces a multi-stage rule-chaining framework that combines symbolic, structural, and conceptual solvers to tackle the Abstraction and Reasoning Corpus (ARC). The architecture achieves over 95 percent accuracy, bridging symbolic reasoning and pattern synthesis to advance transparent machine generalization.
In this solo finale to his three-part series following two weeks in China, Nathan delivers a net assessment of Chinese AI capabilities and constraints ahead of dedicated US-China bilateral AI talks.…
From the tokenization of real-world assets to funding research on anti-aging therapeutics, Coinbase co-founder and CEO Brian Armstrong is focused on solving meta-problems for the benefit of society.…
As reasoning models consume more tokens and AI systems become more expensive to run, understanding what those tokens actually buy is becoming increasingly important. In this episode, Stanford…
Could slowing AI development make superintelligence safer? Daniel Kokotajlo and Thomas Larsen of the AI Futures Project join Tim Scarfe to examine AI 2040: Plan A, a proposal to buy time before AI…
This highlights compilation from AI in the AM captures a landmark week shaped by the releases of Anthropic's Fable 5.1 and OpenAI's GPT-6 Astra alongside new revelations from the OpenAI–Hugging Face…