Agents drive code, enterprise; Regulation builds

Anthropic / Claude ecosystem

Claude Code Plugins Get Official Directory: Anthropic Flags Unverified MCP Risks

Anthropic launched an official, curated GitHub repository for Claude Code plugins, implementing a two-tier verification system that includes automated screening and an optional 'Anthropic Verified' badge. The initiative also explicitly warns users about the risks associated with unverified third-party software, aiming to enhance security and trust in its AI coding agent ecosystem.

Frontier model providers

Elon Musk Confirms Grok V9 Medium Training Completion

xAI has completed the training of its 1.5-trillion-parameter Grok V9 Medium model, with a strong focus on coding performance. The model is expected to launch publicly within weeks, aiming to compete directly in the rapidly evolving AI developer tools market.

Google CodeMender: The AI Agent That Patches Your Code Before You File a Ticket | byteiota

Google has launched CodeMender, an autonomous code security agent now available to external developers. In internal testing, CodeMender has upstreamed 72 verified security fixes by performing root-cause analysis and generating automated patches at scale.

AI developer tooling & infrastructure

How I Indexed 2,000 Claude Code Skills (And What the Install Data Says About AI Coding in 2026) - DEV Community

A new data-driven index by orangebot.ai cataloged over 2,000 Claude Code skills, ranking them by install volume. The analysis reveals that the 'format won,' indicating that all major AI coding agents now support loadable context packages, fundamentally changing how developers discover and adopt coding tools in 2026.

Cloud & platform providers

Announcing Region Expansion of P5.48xl instances on SageMaker Studio notebooks - AWS

AWS expanded the availability of its P5.48xl GPU instances, powered by NVIDIA H100, to seven new regions. This expansion aims to facilitate faster machine learning training and large language model deployment, offering up to four times performance gains for compute-intensive workloads.

AI policy, regulation & governance

UL AI Failure Testing for CNC Systems – Effective 2026

Underwriters Laboratories (UL) has codified new AI-specific failure modes, including training data contamination, adversarial injection, and edge-case misclassification, as mandatory design constraints for CNC control systems. This marks the first major standardization effort to treat AI-enabled motion control as safety-critical.

Massachusetts Bans Weaponized Robots: Drones, AI, and the Future of Warfare (2026)

The Massachusetts House of Representatives has passed sweeping legislation that bans the manufacturing, modification, sale, possession, or operation of weaponized robots within the state. The bill includes exceptions for military and law enforcement use under strict warrant requirements.

Trump’s AI Deregulation Push Signals High-Stakes Battle Over Innovation and Government Control - Tallwire

The Trump administration is prioritizing AI deregulation and advocating for federal preemption over state laws to maintain US competitiveness against China, which is likely to delay comprehensive oversight mechanisms.

Australian media organizations, including News Corp, Nine, ABC, and various industry guilds, are collectively pushing back against the federal government's consideration of relaxing copyright protections. This comes amidst negotiations with AI companies seeking investment commitments in Australia.

Australia and the UK sign AI cooperation agreement on safety and security - Telecompaper

The Australian and UK governments have formally established bilateral cooperation on AI safety and security through a new Memorandum of Understanding. This agreement will facilitate collaboration between their respective AI institutes on critical governance and risk mitigation efforts.

APRA AI Risk Warning: Banks Face Deep-Dive Reviews in 2026

The Australian Prudential Regulation Authority (APRA) has formally warned that AI deployment across Australia's $9.8 trillion in supervised assets is outpacing governance maturity. APRA has initiated enforcement actions against documented failures in model validation and board oversight, signaling deep-dive reviews for banks in 2026.

The AI security gap nobody wants to admit is already here

A leak of Claude Code blueprints exposed internal permission enforcement and sandboxing logic, revealing potential vulnerabilities that attackers could exploit to craft malicious repositories. This incident highlights a growing asymmetry where AI-assisted attacks are outpacing security defenses.

Pope Leo warns of AI fueling warfare in first major theological document | CNN

Pope Leo XIV has issued the first papal encyclical on artificial intelligence, titled 'Magnifica Humanitas.' The document establishes a Catholic social teaching framework for AI governance, ethical considerations in warfare, and the preservation of human dignity in an AI-driven world.

Industry & market moves

xAI Adds $2.8B in Gas Turbines as SpaceX Pitches Space Solar

xAI is investing $2.8 billion in natural gas turbines to expand its AI compute infrastructure, despite facing a pending Clean Air Act lawsuit related to 27 unpermitted Colossus 2 turbines in Mississippi. This move underscores its aggressive strategy to scale computing power for AI development.

SAP To Buy Dremio And Prior Labs To ‘Lead’ Agentic And AI Models - Germany

SAP has acquired Dremio and Prior Labs with a substantial investment of $1.1 billion. This strategic move aims to advance SAP's enterprise agentic AI capabilities by integrating data lakehouse technologies and tabular foundation models into its offerings.

Samsung SDS and KKR Launch 'Startech AI' in New York, Formalizing Global AI Infrastructure Alliance | DailyAlpha

Samsung SDS and KKR have formalized their AI infrastructure alliance by launching 'Startech AI,' a dedicated joint venture entity in New York. This partnership exemplifies the growing trend of 'Big Tech + Private Equity' financing models for scaling AI computing resources globally.

Expert.ai and Fincons Group Join Forces to Bring Neuro-Symbolic AI to Data-Driven Businesses

Expert.ai and Fincons Group have expanded their partnership to integrate neuro-symbolic AI and multi-agent architectures into offerings for regulated enterprises. This collaboration aims to address complex governance and knowledge management challenges inherent in large-scale AI adoption.

NEURA Robotics partners with Bosch to advance German-made robotics - Germany

NEURA Robotics and Bosch have formed a strategic partnership to advance German-made humanoid robotics. The collaboration will focus on joint real-world data collection and co-development of AI software to enhance the capabilities of physical AI systems.

Outcomes4Me: AI-Driven Cancer Care Company Acquires Geno.Me And Realyze Intelligence To Expand Clinical Trial Access

Outcomes4Me, an AI-driven cancer care company, has acquired Geno.Me and Realyze Intelligence. This acquisition aims to combine clinical trial matching and patient data access technologies to accelerate patient enrollment in oncology research.

AI product & feature launches

SAP Unveils EU AI Cloud - Germany

SAP has launched its EU AI Cloud, a sovereign AI and cloud offering tailored for European enterprises, emphasizing data residency compliance. The platform integrates models from partners like Cohere North into the SAP Business Technology Platform.

Fujitsu develops self-evolving multi-AI agent technology that learns and adapts to business operations | Fujitsu Global

Fujitsu has developed a self-evolving multi-AI agent technology designed to autonomously learn from execution results and human feedback. This system continuously improves business-specific LLMs and document search systems without requiring constant expert intervention.

Cerebras Hits 981 Tokens/sec on 1T-Parameter Kimi K2.6, Claims 6.7 GPU Cloud Speedup - DEV Community

Cerebras Systems claims a 6.7x speedup over GPU cloud inference on its 1-trillion-parameter Kimi K2.6 model, achieving 981 tokens/second. The company attributes this performance advantage to its wafer-scale architecture, which eliminates multi-GPU communication bottlenecks.

China is deploying the first home cleaning humanoid robot butlers - Fast Company

GigaAI claims its SeeLight S1 is the first general-purpose humanoid robot designed for home deployment, with 100 pilot units currently rolling out to employee homes. Commercial availability for this home cleaning robot is planned for June 2027.

Workday launches AI agents for IT support & travel

Workday has expanded its Sana AI platform by launching new AI agents specifically for IT support and travel management. These agents automate workflows such as employee onboarding, access changes, and expense processing directly within existing enterprise systems.

Research with immediate practical relevance

Nature and DeepMind forecast AI–human collaboration on Fields Medal by 2030 - DongA Science

A Nature article based on expert forecasts predicts that AI and humans could jointly receive the Fields Medal, mathematics' highest honor, by 2030. This projection is driven by AI systems already making substantive contributions to complex mathematical proof strategies, moving beyond mere calculation to deeper conceptual understanding.

AI Enters the Quantum Materials Lab and Builds a Graphene Transistor

Researchers have developed the first autonomous AI system, named Qumus, capable of independently fabricating complex quantum nanodevices, including a graphene field-effect transistor. This was achieved through robotic manipulation and multi-agent reasoning, marking a significant step in automated scientific discovery.

Tag-teaming turbulence: Frontier Supercomputer trains AI to model cosmic storms - Technology Org

A research team at Argonne National Laboratory used the Frontier Supercomputer to train the first AI model capable of faithfully capturing magnetic turbulence in plasma at extreme conditions. By coupling physics-informed neural operators with generative diffusion, the model reduced prediction errors by over half.

Carnegie Mellon and Cleveland Clinic develop AI system that reads cardiac MRI scans without labeled training data

Carnegie Mellon University and Cleveland Clinic have developed CMR-CLIP, an AI system that interprets cardiac MRI scans directly from radiology reports without requiring manual data labeling. Published in Nature Communications, the system achieved 99% accuracy on some conditions and outperformed general models by 35%.