The day's AI news, briefed
The Daily Prompt
Our AI editor reads each day's AI and robotics stories, pulls out why they matter, and links every original source — so you can keep up in minutes.
Join over 250,000 subscribers getting the latest AI news
Latest AI & Robotics News Today
Programmable World Model
The Programmable World Model framework enables developers to create interactive games with persistent environments, enhancing realism and user control in virtual experiences.
- The Programmable World Model separates state evolution from video generation using rules and 3D bounding boxes.
- It allows agents to translate natural-language instructions into executable programs for controlling entities and interactions.
- CombatStateBench, a new benchmark, evaluates the model's performance, achieving notable accuracy in state management.
- This framework supports coherent generations of long-term interactions in games.
The Programmable World Model represents a significant advancement in creating realistic and interactive video environments by maintaining a persistent world state. This technology aims to enhance user engagement through better control and interaction within virtual spaces. Observers should watch for developments in practical applications and gaming integrations.
Read the original ↗
Show-Harness: Just a VLM Agent Can Play Robots
The development of Show-Harness enables broader applications of vision-language models (VLMs) in robotics, allowing for more efficient control strategies without the need for extensive training.
- Show-Harness connects vision-language models to robot control through discrete semantic actions, enhancing performance and adaptability.
- It allows VLMs to control robots in a zero-shot manner without requiring additional costly training.
- The GUI Manipulation Interface (GUMI) extends these capabilities to graphical user interfaces, making control simpler for human operators.
Show-Harness is an innovative framework that enhances the control of robots by linking them to vision-language models, which are designed to understand and generate language related to visual data. Traditionally, deploying these models in robotics posed challenges, but this new approach shows significant promise for integrating VLMs into various robotic applications. Future developments will likely focus on refining these interfaces and exploring their broader impact in robotics.
Read the original ↗
What Is MLOps? How Teams Build, Deploy, and Monitor Machine Learning Systems
Understanding MLOps is crucial as its principles are essential for effective machine learning systems, impacting performance, governance, and usability in production environments.
- MLOps encompasses engineering and governance for building, deploying, and maintaining machine learning systems in production environments.
- It aims to differentiate between MLOps and DevOps, emphasizing the importance of data and model lifecycle management.
- Key stages in MLOps include versioning data and models, automating training, and monitoring system behavior for effective governance.
- Practical MLOps implementation involves managing inputs, states, and evidence at each stage to ensure robust machine learning outputs.
MLOps, short for Machine Learning Operations, combines software engineering and machine learning principles to streamline production processes. It distinguishes itself from DevOps by emphasizing the unique challenges of managing machine learning models and data. As AI technologies evolve, effective MLOps practices are increasingly necessary to handle complexity and ensure reliability in AI systems.
Read the original ↗
CoreWeave Puts Field Engineers Inside Customer Teams for Physical AI
CoreWeave's new service integrates specialized engineers directly into client teams, enhancing collaboration and improving AI solutions across various engineering fields, which could significantly advance industrial AI applications.
- CoreWeave launched Physical AI Field Engineering on September 10, 2026, aiming to integrate its engineers with customer teams.
- This service enables collaboration on building and validating AI models from existing customer data, improving engineering processes.
- The offering has already been applied in over 100 projects across automotive, aerospace, and robotics industries.
- Key customers include Aston Martin Aramco Formula One Team and Nissan, which have reported successful outcomes from this integrated approach.
- Engagements begin with workshops to align on workflows and establish priorities before model development begins.
CoreWeave provides cloud computing solutions tailored for AI applications. The new Physical AI Field Engineering service leverages their acquired expertise from Monolith AI to bridge the gap between domain expertise and applied AI. Future developments may focus on enhancing the integration of AI in engineering workflows and expanding partnerships with engineering firms.
Read the original ↗
Legal Has Moved from Reviewing AI Decisions to Shaping Them
This shift in the legal landscape emphasizes the growing importance of AI literacy among General Counsels, essential for strategically managing legal risks in fast-evolving technology environments.
- Legal teams are increasingly expected to transform from reactive advisors to proactive business partners, especially in the context of AI.
- Approximately 66% of office professionals are reportedly using AI tools they believe are prohibited, highlighting the prevalence of shadow AI in organizations.
- The role of General Counsels is expanding to include technology governance, requiring them to understand both legal and business implications of AI adoption.
Generative AI has been in use for four years, yet most companies have not standardized its application. The evolving landscape requires legal teams to adapt quickly to technology's influence on business and compliance.
Read the original ↗
Ant International, Visa, Mastercard Align on AI Agent Verification Rules
The collaboration among Ant International, Visa, and Mastercard aims to create a standardized framework for verifying AI agents, which could enhance security and enable significant growth in global commerce.
- Ant International, Visa, and Mastercard announced their collaboration on a Know-Your-Agent (KYA) interoperability framework for agentic commerce on September 10, 2026.
- The KYA framework aims to streamline agent onboarding and identification across payment networks, supporting a projected $3 trillion to $5 trillion in global consumer commerce by 2030.
- This initiative will standardize trust signals across payment ecosystems while maintaining separate verification processes for each network.
Ant International, Visa, and Mastercard are leading companies in the payment processing industry, exploring how interoperability in AI agent verification can support a new wave of commerce. The collaboration builds on existing protocols each company has developed for AI agents. Observers should watch for developments in agentic commerce as this framework is implemented.
Read the original ↗
D-Matrix Connects Raptor XPUs to NVIDIA AI Factories via NVLink Fusion
This collaboration enables d-Matrix to integrate its AI inference technology within a major AI infrastructure, potentially enhancing performance and efficiency in AI applications across industries.
- d-Matrix has announced a collaboration with NVIDIA to integrate its Raptor inference XPUs into NVIDIA's MGX rack architecture.
- The initial integrated availability of these XPUs is expected in the fourth quarter of 2027.
- The partnership will allow high-performance, low-latency AI services for labs and cloud environments using NVLink Fusion technology.
- d-Matrix's Raptor is designed for optimized memory usage, incorporating a unique 3D DRAM stacking approach.
- NVIDIA's AI infrastructure will support efficient scalability with integrated systems and partners in the NVLink Fusion ecosystem.
d-Matrix is a chipmaker specializing in AI inference technology, while NVIDIA is a leading provider of AI hardware and software solutions. This partnership is part of a broader trend towards integrating custom silicon with advanced AI infrastructure to meet increasing demand for inference capabilities. Future developments to watch include the rollout of Raptor XPUs and their performance in real-world applications.
Read the original ↗
Salesforce Unveils Six-Capability Trusted AI Harness for Enterprises
Salesforce's new AI harness aims to integrate multiple enterprise capabilities efficiently, shaping how organizations can manage interactions and enhance customer service with AI at scale.
- Salesforce introduced the Trusted Enterprise AI Harness, which features a composable architecture with six trusted capabilities designed for enterprise needs.
- The harness aims to equip AI agents with the ability to reliably and securely fulfill complex business tasks across different systems in real-time.
- A new AI Control Plane will provide companies centralized management and visibility over their AI agents and capabilities throughout the enterprise.
- The six capabilities include Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models, each serving distinct roles in operationalizing AI.
Salesforce is a leading customer relationship management (CRM) platform that integrates various business processes and data. The Trusted Enterprise AI Harness builds on efforts to enhance AI usability by connecting different enterprise functions securely and efficiently. As Salesforce rolls out these capabilities, organizations will need to consider how these tools can be integrated into existing frameworks.
Read the original ↗
OpenAI Brings Top AI Doomer Onto Its Board
The addition of Paul Christiano to OpenAI's board highlights ongoing concerns about AI safety, emphasizing the need for oversight as AI systems grow more powerful and complex.
- OpenAI has appointed Paul Christiano, known for his concerns about AI safety, to its board of directors.
- Christiano will join the Safety and Security Committee, which decides whether to release new AI models.
- He warned that rapid advancements in AI could lead to loss of human control, with potential irreversible consequences.
OpenAI is a leading AI research lab known for developing powerful AI models, including chatbots. Christiano previously worked at OpenAI and founded the Alignment Research Center, focusing on AI safety issues. His board appointment comes amid rising scrutiny regarding AI technology's safety and ethical implications.
Read the original ↗
Clearview AI Is Testing an AI Tool That Would Let Cops Unearth Your Life Online
The development of InquiryIQ by Clearview AI could expedite police investigations, but raises significant concerns regarding privacy and the potential for expanded surveillance.
- Clearview AI is testing an experimental AI tool named InquiryIQ to assist law enforcement in uncovering detailed online information about identified individuals.
- The tool is designed to automate the compilation of profiles that include personal data like employers, associations, and online activity.
- InquiryIQ's interface suggests inputting demographic data to enhance search results, but its application and implications remain unclear.
- Experts warn that such rapid investigation tools could result in less oversight and increase the risk of unjust scrutiny towards individuals.
- Clearview AI currently claims InquiryIQ is a prototype and has not been offered to law enforcement agencies yet.
Clearview AI is known for its controversial face-recognition technology that extracts images from the internet for law enforcement use. Founded in 2017, it has grown its image database significantly amidst legal challenges and public scrutiny regarding privacy. Continuous development of tools like InquiryIQ reflects the ongoing blending of AI technology with policing practices.
Read the original ↗
Sorry, Apple, but the iPhone Duo Wasn’t the Smart Display I’m Waiting For
The delay in releasing a dedicated Apple smart display may hinder the company's ability to compete in the growing smart home market dominated by established players.
- Apple recently showcased its new foldable iPhone Duo as a personal hub for AI but did not unveil the anticipated smart home display.
- Reports indicated that the smart display would have features like facial recognition and multiple installation options but were not confirmed during the fall event.
- The iPhone Duo, priced at $1,999, is intended to serve similar functions as a smart display but lacks essential features and capabilities.
- Apple's delay in releasing a smart display risks losing market share as competitors like Amazon and Google have established dominance in this area.
Apple has not yet released a dedicated smart home display, which many consumers anticipated for the fall event. The company has been focusing on integrating AI features into existing products, such as the iPhone, while facing challenges in competing with established smart home devices.
Read the original ↗
NYU-DRP AI Model Predicts Five-Year Breast Cancer Risk From 3D Mammograms
The NYU-DRP model enhances breast cancer risk prediction accuracy, potentially allowing for more personalized screening and improved patient outcomes in breast cancer detection.
- Researchers at NYU Langone Health have developed a deep learning tool, NYU-DRP, to estimate a woman’s five-year breast cancer risk from 3D mammograms.
- The model was validated using 313,335 mammogram examinations from over 161,000 women and shows improved risk prediction compared to existing models.
- The study found that using longitudinal mammograms provided risk information not captured by single mammograms or breast density alone.
NYU Langone Health's Perlmutter Cancer Center has created an AI model to improve breast cancer risk prediction through repeated 3D mammograms, which are increasingly used for screening. The model aims to inform personalized screening strategies and will be tested further in diverse populations.
Read the original ↗
Fujitsu Signs New Palantir AIP Agreement, Becomes Global FDE Partner
This partnership enhances enterprise AI capabilities, addressing increased demand for data sovereignty and operational efficiency, which could significantly impact AI application's reliability and performance in businesses.
- Fujitsu and Palantir Technologies renewed their strategic partnership on September 10, 2026, focusing on Palantir AIP and Foundry.
- Fujitsu will serve as a Global Forward Deployed Engineering (FDE) Partner, investing in capabilities to support AI transformation globally.
- The collaboration aims to help customers achieve greater control over data and accelerate decision-making processes.
- A supply chain solution implemented for a leading manufacturer resulted in over $10 million savings and doubled operational productivity within a year.
Fujitsu has partnered with Palantir since 2020, focusing on applying AI solutions to improve business outcomes. The renewed agreement presents a commitment to enhancing Forward Deployed Engineering capabilities important for implementing AI solutions. Companies are working together to assist organizations in Japan and internationally in adopting these new technologies.
Read the original ↗
Mistral Integrates Models With Cloudera for Sovereign Enterprise AI
This partnership enhances data security and control for enterprises, addressing the growing demand for sovereign AI solutions where organizations retain ownership of their data and models.
- Mistral AI has partnered with Cloudera to integrate its AI models with Cloudera's hybrid data platform for enterprise customers.
- The integration allows for AI model deployment in fully controlled environments, including private and air-gapped installations.
- Enterprises can train Mistral’s AI models on proprietary data while retaining ownership of both the data and the resulting AI intelligence.
Mistral AI focuses on sovereign AI, defined as retaining complete customer control over data and models. Cloudera provides a data platform that enables enterprises to manage and utilize large datasets effectively. This partnership aims to enhance the ability of businesses to leverage AI while maintaining data sovereignty.
Read the original ↗
In AI, the Walls Aren’t Real: Rethinking Data & Safety
Addressing the shortcomings of AI concerning data utilization and understanding can significantly enhance the development of more effective and empathetic AI systems.
- Current AI systems struggle with data limitations, financial scaling, rogue behavior, and catastrophic forgetting when altered.
- Experts argue that AI is not running out of data, as much of the available data has not been fully leveraged.
- Enhancing AI can be achieved by using high-quality human datasets more effectively, rather than generating synthetic data.
AI technology is increasingly employed across various sectors, yet it faces challenges in optimizing data use and achieving emotional intelligence. The article critiques existing AI training methods and suggests improvements for future systems. Attention will be focused on how AI can better integrate lived human experiences to enhance its functionality.
Read the original ↗
Task-Aligned vs. Human-Aligned: Why AI’s Next Benchmark Should Be Us
As AI continues to advance rapidly, understanding the impacts of these technologies on human decision-making and critical thinking is crucial for responsible integration into everyday life.
- Frontier AI models improved by approximately 30 percentage points on Humanity’s Last Exam, a benchmark designed to challenge AI capabilities.
- Task-aligned AI focuses on completing specific requests, while human-aligned AI assesses the impact on user capabilities after interactions.
- Current benchmarks often overlook the subjective elements of user experience, potentially harming critical thinking and self-reliance among users.
The article discusses the distinction between task-aligned and human-aligned AI, emphasizing the need for benchmarking that evaluates the long-term impact on users. It argues that better tools and evaluation methods are necessary to ensure AI supports human decision-making rather than undermining it.
Read the original ↗
AI Is Not Taking Our Jobs. But Energy Costs Might.
The energy costs driven by AI demand threaten to stifle economic growth and job creation in the UK, as companies struggle with unpredictable energy prices and lengthy connections to the power grid.
- The share of UK businesses using AI has nearly tripled from 12% to 35% since late 2023, but headcount remains unchanged.
- Global data center electricity demand is expected to double by 2030, placing a strain on energy generation and infrastructure.
- The UK grid connection queue has surged from 41 GW to 125 GW, with significant delays impacting data center projects and economic ambitions.
- Rising energy prices could deter companies from expanding, ultimately affecting job creation and competitiveness.
- Policymakers are urged to address energy pricing and grid connection issues to facilitate growth in industries reliant on AI.
AI technology is increasingly adopted by UK businesses, influencing how tasks are performed rather than eliminating jobs. However, energy supply and pricing issues stand as significant barriers to economic growth. Moving forward, tackling these energy challenges will be critical for maintaining competitiveness in the AI-driven economy.
Read the original ↗
JD.com expands physical AI in logistics with 3 million robots
JD.com's expansion of robotic logistics reflects a significant escalation in automation within the supply chain industry, potentially transforming delivery efficiency and operational costs in the ecommerce sector.
- JD.com plans to procure 3 million robots, 1 million autonomous vehicles, and 100,000 drones over the next five years as part of its Physical AI Acceleration Plan.
- The company unveiled its industrial Wolf Robot series, intended for various logistics tasks including warehousing and transport, at JDDiscovery 2026.
- JD Logistics has deployed its automated warehousing system across more than 30 warehouses in China and in the UK and Germany as of June 30.
- JD has expanded its autonomous vehicle network into night-time delivery routes in Shenzhen, enhancing operational efficiency and reach.
- The firm is also developing a large AI computing cluster with 100,000 GPUs for training and inference tasks.
JD.com is a major Chinese ecommerce company that is enhancing its logistics services through automation and robotics. The company's previous initiatives include deploying automated systems in warehouses and expanding its drone delivery network. Observers should watch how these developments impact competition with Alibaba and Meituan in the logistics space.
Read the original ↗
NVIDIA and Palantir Announce Sovereign AI Stack for Supply Chains
This collaboration between NVIDIA and Palantir introduces a new AI solution that aims to enhance the efficiency and responsiveness of supply chain operations across various industries.
- NVIDIA and Palantir Technologies announced a partnership to develop a sovereign AI stack for supply chains on September 10, 2026.
- The stack will initially be deployed within NVIDIA's supply chain and includes features for materials allocation decisions while protecting customer data ownership.
- Key industries such as agriculture, manufacturing, and technology can apply this AI stack to improve their supply chains.
NVIDIA is a leader in graphics processing units and AI computing technologies, while Palantir specializes in data analytics and software for data integration. Previous efforts in supply chain management by NVIDIA have highlighted the coordination complexities involved, and this new system aims to streamline those processes. Upcoming developments include applying learnings from the NVIDIA deployment to other sectors like healthcare and automotive.
Read the original ↗
Cardinal Program Brings Free AI-Driven Security Testing To US Utilities
The Cardinal Program aims to enhance the cybersecurity of U.S. critical infrastructure, which is increasingly threatened by sophisticated cyber attacks leveraging AI technology.
- Method Security and Palantir Technologies jointly launched the Cardinal Program on September 10, 2026, offering free AI-driven security assessments to U.S. utilities and municipalities.
- The program provides continuous autonomous security assessments at no cost, aimed at identifying vulnerabilities before they can be exploited by cyber adversaries.
- Participants can set their own security assessment parameters, ensuring that all actions adhere to pre-defined engagement rules and human oversight is enforced.
The Cardinal Program is designed to help U.S. organizations improve their cybersecurity posture amid rising threats, particularly those posed by AI-enabled attackers. Method Security specializes in autonomous cyber systems, and Palantir provides supporting infrastructure and technology solutions. Observers should watch for enrollment trends and how the initiative may reshape cybersecurity practices among public services.
Read the original ↗