AI News & Research Breakthroughs
Notable AI breakthroughs, research, and tools — the developments every business owner should have on their radar.
Significant AI news, research breakthroughs, new models, tools, and applications that do not fit a single category.
Waymo Publishes Breakthrough Human Collision Avoidance Model With TU Delft in Nature Communications
Waymo has published joint research with TU Delft in Nature Communications introducing a new active inference framework — called NIEON — that models how humans naturally avoid crashes. The model is designed to improve Waymo's ability to predict and respond to human driver behavior in complex traffic situations. The publication is part of Waymo's broader push to establish itself as the 'world's most trusted driver' by grounding its systems in peer-reviewed safety science.
See the business automation angle on this story — free with signup.
TraceGains Launches Formula AI to Embed Intelligence Directly Into Food Product Development
TraceGains has released Formula AI, a tool that integrates AI directly into the food R&D process by combining formulation tools, supplier and ingredient intelligence, compliance data, and collaborative workflows into a single environment for food scientists. The platform uses a specialized knowledge layer built on real-world industry data, formulation rule sets, and scientific workflows to help developers create products faster with fewer iterations. It aims to reduce development costs and accelerate time-to-market for food and beverage manufacturers.
See the business automation angle on this story — free with signup.
Peer-Reviewed Study Finds AI Models Absorb and Replicate Antisemitic Tropes From Training Data
A study published in American Psychologist found that large language models including ChatGPT and DeepSeek replicate historical antisemitic stereotypes absorbed from their human-generated training data, even with guardrails in place. Researchers stripped identifying information from biographies and had both AI models and 378 human participants rate the subjects, finding the models reproduced biases that humans also showed but to a measurable degree. The study adds to a growing body of evidence that LLM bias reflects and can amplify pre-existing social prejudices.
See the business automation angle on this story — free with signup.
Decart Launches Oasis 3 World Model to Generate More Realistic Robot Training Environments
AI research lab Decart has launched Oasis 3, a video output world model designed to produce more physically realistic synthetic environments for training robots and autonomous vehicles. The model aims to bridge the gap between current rigid physics engines — which struggle to capture real-world nuance — and the unpredictable environments robots must ultimately operate in. More realistic simulated training data is expected to accelerate robot deployment timelines.
See the business automation angle on this story — free with signup.
Pi Security Startup Raises $35M to Use Agentic AI for Automated Vulnerability Remediation
San Francisco-based Pi, founded by former Microsoft and Tesla security leaders, has raised $35 million to scale an agentic AI platform that can reduce vulnerability triage time by up to 80% by automatically identifying and fixing software security flaws from design through deployment. The company describes its technology as a 'security brain' that combines institutional knowledge of a company's software history with AI reasoning to distinguish real threats from false alarms. Customers reportedly include a leading AI lab, cybersecurity firms, and a major social network.
See the business automation angle on this story — free with signup.
Microsoft Cutting Staff in Azure Cloud Unit as AI Restructuring Continues
Microsoft has announced workforce reductions in its Azure cloud division, the latest in a series of restructuring moves as the company realigns resources toward AI-driven infrastructure and services. The cuts reflect a broader industry pattern where traditional cloud operations roles are being reduced as AI automation takes over routine tasks. Microsoft has not disclosed the full scope of the layoffs.
See the business automation angle on this story — free with signup.
Researchers Build Self-Replicating AI Worm That Spreads Across Networks Using Local Open-Weight LLMs
University of Toronto researchers demonstrated a proof-of-concept AI-powered computer worm that uses a locally hosted open-weight large language model to autonomously reason through a network, generate custom attack strategies per target, and replicate itself — all without human input or any commercial AI service. In 15 isolated test runs on a 33-host vulnerable network, the worm identified an average of 31 vulnerabilities and successfully replicated to 62% of the network over seven days. The research highlights a new category of AI-enabled cyberattack that operates entirely on-premises, bypassing API-based safety guardrails.
See the business automation angle on this story — free with signup.
JPMorgan Chase Plans to Deploy Long-Running Autonomous AI Agents Across Banking Operations This Year
JPMorgan Chase revealed exclusively to CNBC that it plans to deploy AI agents capable of operating autonomously for hours at a time later in 2026, a significant step beyond its current shorter-running AI tools. The bank said existing AI deployments have already driven a 20% increase in private banking gross sales, and believes autonomous agents could allow individual bankers to expand client coverage by up to 50%. JPMorgan's focus has shifted from model intelligence to agent endurance — how long AI systems can operate reliably without human intervention.
See the business automation angle on this story — free with signup.
Google Pays SpaceX $30 Billion for AI Compute as Infrastructure Costs Soar
Google has agreed to pay SpaceX approximately $30 billion for AI compute capacity, underscoring the extreme demand for processing power to train and run advanced AI models. Separately, NVIDIA reports that its latest AI systems now cost $7.8 million to build, with memory alone comprising 25% of total cost — a 485% increase in memory costs. These figures highlight how the economics of frontier AI development are becoming increasingly concentrated among a small number of infrastructure players.
See the business automation angle on this story — free with signup.
Research Finds AI Boosts Human Creativity, Tested Against 100,000 People
Scientists conducted a large-scale study pitting AI against 100,000 human participants on creativity tasks, finding that AI can measurably enhance human creative output rather than replace it. The research suggests AI works best as a collaborative tool that expands what people can produce rather than operating independently. This adds to a growing body of evidence that human-AI collaboration outperforms either working alone.
See the business automation angle on this story — free with signup.
NVIDIA Launches Nemotron 3 Super, a 120B-Parameter Reasoning Model for Multi-Agent AI
NVIDIA released Nemotron 3 Super, a reasoning model with approximately 120 billion parameters specifically designed for use in multi-agent AI systems. The model is aimed at enterprise automation infrastructure where multiple AI agents need to collaborate on complex tasks. Its release adds a major open-weights option to the competitive landscape alongside models from Anthropic, OpenAI, and Google.
See the business automation angle on this story — free with signup.
Nvidia Releases Nemotron 3 Super: A 120B Parameter Reasoning Model for Multi-Agent AI
Nvidia released Nemotron 3 Super, a 120-billion parameter reasoning model specifically designed for multi-agent AI systems that can coordinate multiple AI models working together on complex tasks. The release is part of a broader wave of new AI tools hitting the market rapidly, with developers already experimenting with how to integrate it into automation workflows. Two additional mystery AI models also appeared on the OpenRouter platform around the same time, fueling speculation about upcoming announcements.
See the business automation angle on this story — free with signup.