NVIDIA Robots Deploying in Poland, Maxinsights Raises $30M

Key Takeaways

• <ul><li>Otto Group and Sereact plan to deploy up to 25 NVIDIA-powered robots at a Poland logistics center</li><li>Sereact's Cortex Physical AI model can handle items without prior specific training</li><li>Maxinsights raised $30 million for physical AI training data, collecting over 2 million hours of real-world human activity</li><li>Sam Altman urged companies to prioritize AI security threats before pursuing productivity gains</li><li>Intel and SAP demonstrated a proof of concept linking enterprise workflows to autonomous warehouse robots</li><li>China's CXMT produces HBM3E chips with only 20-30% yield compared to SK Hynix's ~70%</li><li>FoG framework enables smaller AI agents (3B-4B) to outperform larger models in evidence retrieval</li><li>Allplan Max 2027 brings AI tools across engineering platforms including Allplan, Civil, and SCIA</li><li>AI-related securities lawsuits surged to 15 filings in first half of 2026, with new diversion theory targeting Microsoft, ZoomInfo, and monday.com</li><li>Microsoft faces claims about $72.4 billion in AI capital expenditures diverting Azure capacity to Copilot</li></ul>

Otto Group and Sereact deploy NVIDIA-powered robots for returns

Otto Group and Sereact will deploy up to 25 AI-controlled dual-arm robots at an Otto Group logistics center in Poland. The robots run on Cortex, Sereact's Physical AI model, which can recognize and handle items it has never seen before. They are trained using NVIDIA Isaac Sim, NVIDIA Isaac Lab and NVIDIA Cosmos. The project will run in two phases, starting with ten systems and adding 15 more later.

AI-powered robots aim to automate retail returns processing

Otto Group and Sereact are working together to deploy up to 25 NVIDIA-powered AI-controlled robots in a retail logistics center. The robots use Sereact's Cortex Physical AI to process returned items without needing prior specific training. NVIDIA's simulation and testing systems help validate the technology. The partners plan a two-phase deployment to test whether the system can work at large scale.

Sam Altman urges companies to prioritize AI security threats

OpenAI CEO Sam Altman said companies should focus on understanding AI security threats before using AI for productivity. He said AI systems are finding many serious software vulnerabilities that need to be fixed. He noted that new monitoring services are needed because it is too hard to patch every bug one by one. His comments came days after AI agents helped uncover serious flaws in PHP.

Altman says AI security must come before productivity

OpenAI CEO Sam Altman urged companies to make understanding AI security threats a top priority. He said he was shocked by how many software flaws AI systems are finding. He recommended that businesses focus on mitigating these risks before using AI tools like Codex to become more productive. He explained that new methods are needed to monitor traffic and stop flaws from being exploited.

Intel and SAP demonstrate Physical AI for warehouse robots

Intel and SAP showed a proof of concept that connects enterprise workflows to autonomous robots. Requests from SAP Extended Warehouse Management are sent to Intel-powered edge AI robots that transport materials and report back. The system creates a closed loop between business systems and physical action. The demonstration showed reduced transport distance and less repetitive manual handling in warehouse operations.

ARC Terminal returns with AI powered crypto trading tool

ARC Terminal is rebuilding as an AI powered crypto trading platform. It aims to combine market analysis, opportunity finding, and trade execution into one interface. The team wants AI agents that understand user instructions and remember important context. The project must prove it can execute trades reliably and earn user trust.

China pushes advanced memory chips despite 70 percent failure rate

China's largest DRAM maker CXMT is producing advanced HBM3E memory chips for AI processors despite very low yields. Only about 20 to 30 percent of its chips meet specifications, while competitor SK Hynix reaches around 70 percent yields. CXMT's chips also perform closer to SK Hynix's older HBM3 products. China faces US export restrictions on advanced memory and chipmaking equipment. The government is pushing homegrown AI alternatives, which requires a strong domestic supply of advanced memory chips.

FoG framework trains AI agents to retrieve supporting evidence

Researchers created FoG, a framework that trains knowledge graph agents to retrieve the evidence behind their answers. Existing agents often give correct answers without showing their supporting data. FoG uses evidence grounded synthesis and a process reward called ECPR to track supporting triples. Experiments show that 3B to 4B agents using FoG outperform larger models in accuracy and evidence coverage.

Allplan Max brings AI tools to engineering workflows

Allplan Max 2027 is a new AI platform designed for engineering teams. It works across Allplan, Allplan Civil, Frilo, SCIA, and SDS2 to automate tasks like searching, reporting, and drawing cleanup. Features include AI Formula Writer, Workflow Builder, AI BIM Analyst, and AI Drafter. The company says engineers keep responsibility for design intent and final decisions.

Maxinsights raises 30 million for physical AI training data

Maxinsights raised 30 million to build a training data platform for physical AI. The company has collected over 2 million hours of real world human activity across more than 30 customers. It launched MaxLoop to review footage and MaxGlove to capture touch and grip data. The team includes veterans from Waymo, Google, Microsoft, and Meta, and the round was led by PeakXV.

Data-Driven Approach Delivers Actionable Recourse in Machine Learning

Machine learning systems now make important decisions about people's lives. The paper proposes Data-Driven Action Recourse as a new method. This approach aims to give users practical steps to achieve desired outcomes. It moves away from costly human supervision that requires hand-designed cost functions. The goal is to find low-cost actions that help people reach their goals.

AI Diversion Theories Create New Securities Lawsuit Risks

AI-related securities lawsuits have surged from 16 filings last year to 15 in the first half of 2026. Cornerstone reports this year could nearly double previous filings. Plaintiffs use two main theories: companies overstating AI abilities or hiding AI-related risks. Recent cases against Microsoft, ZoomInfo, and monday.com show a new AI diversion theory. This theory claims companies hide how AI shifts resources and customers away from core business. Microsoft faces claims about $72.4 billion in AI capital expenditures diverting Azure capacity to Copilot. ZoomInfo and monday.com face claims about customers building internal AI tools or using AI search instead of their products.

Sources

• Otto Group and Sereact Bring NVIDIA-Powered Physical AI to Returns Processing
• AI Chips Update - AI-Driven Robots Revolutionize Retail Logistics Automation
• Prioritise understanding AI security threats, even before productivity: OpenAI's Sam Altman to cos
• Make AI security priority, not productivity: Altman
• From Task to Fulfillment – Intel and SAP Demonstrate Physical AI and...
• Gm gARC ARC TERMINAL: REBUILDING WITH A NEW AI VISION @TheA
• China Is So Desperate for AI Memory That It’s Producing Chips With a 70% Failure Rate
• Clearing the FoG: Training KG Agents to Retrieve the Evidence Behind Their Answers
• Allplan Max delivers engineering AI across portfolio
• Maxinsights raises $30 million to build training data platform for physical AI
• Follow the Herd: A Data-Driven Approach for Actionable Recourse
• AI Diversion Theories Emerge as New Securities Lawsuit Risk Zone