The news, with the noise removed.

Nvidia, Microsoft, SpaceX, and IBM have formed the Open Secure AI Alliance to develop and share open-source security tools for frontier AI models. The initiative aims to address growing safety concerns by providing defensive tools against potential attacks. Notably absent from the alliance are OpenAI, Google, and Anthropic, raising questions about the landscape of AI security collaboration. The alliance's success will depend on whether these tools can defend against actual attacks in production environments, not just in controlled demonstrations.

The Trump administration has announced the first Genesis Mission grants, allocating $5 billion to support hundreds of AI-driven science projects. The program has been described as comparable to the Manhattan Project, marking a significant shift toward AI in scientific research funding. The success of this initiative will depend on whether the funded projects produce meaningful scientific advances rather than just demonstrating AI's theoretical potential.

Anthropic has released Claude Opus 5, a new AI model that closely matches the capabilities of Claude Fable 5 across various domains. The release follows recent industry discussions on AI safety and represents a significant advancement in Anthropic's model development. The model's true value will be determined by actual user testing and whether it delivers meaningfully better performance in production environments.

OpenAI is rolling out ChatGPT Health to the entire US, enabling users to link their medical records and health data to the chatbot. OpenAI claims its models can now reason at levels surpassing clinician capabilities, marking a significant step in AI-driven healthcare. The success of this rollout will depend on actual user outcomes, liability frameworks, and whether the system can handle real-world medical complexity rather than controlled examples.

Meta is enhancing its AI chatbot with new productivity features, including calendar integration for event planning and daily briefings. The update aims to compete with rivals like Gemini, ChatGPT, and Claude by offering in-depth research capabilities guided by user input. The success of these features will depend on whether they integrate smoothly into existing workflows and provide genuine productivity gains rather than just additional complexity.

Midjourney, known for AI-generated images, has acquired the astrology app Co-Star, expanding into the astrology sector. Co-Star offers personalized horoscopes and astrological insights, marking Midjourney's entry into a new field of AI applications. The strategic logic behind this acquisition will become clearer as the integration progresses and user adoption patterns emerge.

Nvidia, Microsoft, SpaceX, and IBM have launched the Open Secure AI Alliance, a collaboration focused on developing open-source AI security tools. The initiative addresses growing concerns about frontier model safety by creating shared defensive capabilities against potential attacks. Notably absent from the alliance are OpenAI, Google, and Anthropic, though the founding members frame this as an infrastructure play rather than an exclusionary move.

The Trump administration has announced the Genesis Mission, a $5 billion funding initiative for AI-driven science projects. Officials have compared the scale of the effort to the Manhattan Project, marking a significant shift toward tech-centric scientific research and funding priorities.

OpenAI research demonstrates how AI, particularly ChatGPT, is reshaping workplace dynamics by enabling workers across various roles to take on new tasks and responsibilities. The study shows AI expanding job boundaries rather than simply automating existing functions.

Anthropic has released Claude Opus 5, a new AI model that closely matches the capabilities of Fable 5 across many domains. The launch positions Anthropic as a key contender in the ongoing frontier model competition.

OpenAI has rolled out ChatGPT Health to the entire U.S., enabling users to connect medical records for personalized health insights. The company claims its models now demonstrate reasoning capabilities that surpass clinician levels, representing a significant advancement in AI-driven healthcare.

Meta is enhancing its AI chatbot with productivity features including calendar integration and daily briefings, aiming to compete more effectively with rivals like Gemini, ChatGPT, and Claude by offering comprehensive assistant functionality.

NVIDIA has joined the Open Secure AI Alliance alongside Microsoft, SpaceX, IBM, and other organizations to develop open-source AI security tools. The alliance focuses on defense against attacks from frontier models, addressing growing safety concerns as AI systems become more capable. Notably, OpenAI, Google, and Anthropic are not participating in this initial round. The move signals NVIDIA's growing involvement in the security layer of the AI infrastructure stack.

NVIDIA's technical blog outlines six key capabilities required in agent harness architectures to achieve higher model performance. The post emphasizes that effective AI agents require more than model selection; the underlying architecture must efficiently render context and execute tasks. This represents NVIDIA's focus on the software and infrastructure stack that enables agentic AI workloads.

NVIDIA has set a new world record for Mixture of Experts (MoE) pre-training using the GB300 NVL72 system. The achievement demonstrates how MoE architectures are transforming large-scale AI training by optimizing compute efficiency per token while maintaining high model performance. This represents a significant milestone in rack-scale training infrastructure.

NVIDIA's ModelExpress is a new system designed to efficiently distribute large model artifacts and checkpoints. As model sizes continue to grow, transferring these files has become a significant bottleneck. ModelExpress aims to reduce costs and improve speed in moving large datasets across infrastructure, addressing a critical piece of the AI deployment pipeline.

NVIDIA has unveiled the Rubin GPU architecture, designed specifically for the agentic AI era. The architecture supports always-on AI factories capable of producing intelligence at scale, representing a significant evolution from traditional discrete AI training models. Rubin focuses on sustained production workloads rather than one-off training runs.

NVIDIA's Vera CPU introduces Olympus Cores, a new architecture optimized for maximum single-thread performance specifically for agentic AI workloads. The design enhances the CPU's role in executing complex, multi-step agent operations, complementing GPU acceleration with strong single-thread capability. This represents NVIDIA's expansion into full system-level design for AI infrastructure.