OpenAI boosts memory; global AI regulation widens

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

Dreaming: Better memory for a more helpful ChatGPT | OpenAI

OpenAI has launched 'Dreaming V3,' a significantly more capable memory architecture for ChatGPT. This new system automatically synthesizes user context across conversations, allowing the AI to remain current and relevant over extended interactions.

🤖 The AI that solves unsolved problems now works on healthcare, power grids, and industry

Google's AlphaEvolve AI system has been deployed across diverse sectors, including healthcare genomics, power grid optimization, and enterprise processes. The system has demonstrated significant improvements, such as a 30% error reduction in DNA analysis and an 88% improvement in power flow solutions.

AI developer tooling & infrastructure

And Now I've Vibe-Coded a Native Windows 11 App ⭐ - Thurrott.com

AI-assisted CLI-based code generation tools now enable developers to 'vibe-code' native Windows apps end-to-end. These tools leverage grounded knowledge of WinUI3 APIs, allowing for intuitive and rapid development of Windows applications.

Devin Desktop | Devin

Devin Desktop has launched as a unified IDE designed to manage fleets of local and cloud coding agents. This platform allows developers to plan, delegate, review, and ship code directly from their editor, streamlining the agent-driven development workflow.

Harness May 2026 Product Updates: 60+ New Features

Harness shipped over 60 new features in May 2026, enhancing its delivery platform with AI-native capabilities like the MCP Server in Claude Connectors Directory. Other updates include security improvements such as keyless artifact signing and typosquatting detection, alongside reliability features like flaky test detection.

Using the marimo editor's AI features - marimo

The marimo AI-assisted coding editor offers a unique feature: its AI assistant has direct access to variable values and dataframe schemas in memory. This enables the generation of code that is grounded in the live data context, improving accuracy and relevance.

Agent Memory Guard: Middleware to prevent memory poisoning in multi-agent systems · microsoft autogen · Discussion #7792 · GitHub

New open-source middleware, 'Agent Memory Guard,' has been introduced to detect and prevent memory poisoning attacks in multi-agent AI systems. It employs a five-layer defense strategy, including heuristics, embeddings, entropy, encoding, and an LLM-judge, all operating with sub-microsecond latency.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

Top AI leaders, including CEOs from OpenAI, Anthropic, and Google DeepMind, are urging Congress to mandate legal safeguards on synthetic DNA/RNA ordering. Their call aims to prevent the misuse of AI in designing biological weapons.

Tech: Obernolte-Trahan AI proposal is out - Punchbowl News

A bipartisan proposal, the Great American Artificial Intelligence Act, has been introduced in the US House of Representatives. The bill aims to regulate AI by requiring transparency into powerful model risks and preempting some state laws concerning AI.

Supreme Court AI draft rules: No AI in verdicts, bail eligibility

India's Supreme Court AI Committee has proposed draft regulations that would prohibit the use of AI in judicial verdicts and bail eligibility decisions. The rules emphasize that all AI systems must remain subservient to human judicial authority.

NSW Government Bulletin: Managing The Use Of Digital Work Systems In Government Agencies - Employee Rights/ Labour Relations - Australia

The NSW Government has enacted Australia's first express statutory duty specifically addressing digital work systems. This law requires government agencies to assess and manage risks associated with algorithmic work allocation, performance metrics, and worker surveillance.

Australian music industry organizations ARIA and APRA AMCOS have strongly disputed claims by Scott Farquhar that AI model training is legally impossible under Australian copyright law. They assert that comprehensive music licensing agreements could facilitate AI training.

Push to include AI chatbots in Australia's under-16 social media ban

Advocates in Australia are pushing for an expansion of the country's existing under-16 social media ban to include AI chatbots. This move aims to protect minors from potential harms associated with advanced AI interactions.

Industry & market moves

Anthropic and University of Tokyo partner to gauge generative AI use - Nikkei Asia

Anthropic is partnering with the University of Tokyo to develop an index that will measure the impact of generative AI across Japanese business, government, and education sectors. This collaboration aims to provide concrete data on AI adoption and its effects.

xAI pauses hiring for specialists to train Grok chatbot

xAI has paused its specialist hiring for Grok chatbot training, following recent layoffs of 500 generalist tutors. This move signals a potential shift in xAI's AI training strategy amidst increasing regulatory scrutiny.

Alnylam pens $2B deal with Inceptive to transform siRNA design with AI pioneer | Fierce Biotech

Alnylam has partnered with Inceptive in a $2 billion deal to leverage AI-driven sequence modeling for advancing siRNA drug design. This collaboration combines transformer-based models with large-scale biomedical data to optimize drug development.

Nvidia snaps up Kumo AI, a predictive AI startup known for its extreme accuracy - SiliconANGLE

Nvidia has acquired Kumo AI Inc., a predictive AI startup recognized for its highly accurate business forecasting capabilities, powered by Graph Neural Network models accessible via natural language queries.

Andreessen Horowitz and Forerunner are betting $55 million on Town's deeply personal AI assistants | Fortune

Andreessen Horowitz and Forerunner have led a $55 million Series A funding round for Town, a company developing personalized AI assistants. The investment aims to enhance user productivity by giving them back time through distinct AI avatars and personalities.

Cognizant Accelerates Enterprise AI Adoption with Snowflake's Cortex-Powered Intelligent Agents

Cognizant is expanding its collaboration with Snowflake CoCo to deploy production-grade AI agents across enterprise workflows. This initiative aims to reduce time-to-value from weeks to hours, with over 2,250 users and 30+ operationalized use cases already achieved.

Global Objects Announces Strategic Collaboration with Microsoft to Build the Ground-Truth Foundation for Physical AI on Microsoft Azure - Las Vegas Sun News

Global Objects, Inc. has partnered with Microsoft to develop a physics-grounded generative AI world model. This model will be trained on licensed 3D scans rather than internet data, aiming to address a critical bottleneck in Physical AI by providing verifiable, real-world datasets.

AI product & feature launches

Trend Micro to Use Mythos for Cyber Defense Services | Nippon.com

Trend Micro announced it will integrate Anthropic's Claude Mythos model into its cyber defense services, becoming the first Japanese company to publicly disclose access to the advanced AI for enterprise security.

SpaceXAI powers Gopuff's Go AI shopping assistant | Let's Data Science

Gopuff has launched 'Go,' a new Grok-powered AI shopping assistant that combines SpaceXAI's reasoning and generative models with Gopuff's extensive 13-year order dataset and real-time X signals. This integration enables instant delivery services in as fast as 15 minutes.

MWM AI Partners with Google Cloud to Launch New Mobile Agents for App Creators and Entrepreneurs - Jun 4, 2026

MWM AI and Google Cloud have partnered to launch 'AI Mobile Squad,' a coordinated team of three specialized AI agents (Designer, Product Manager, Developer). This squad can transform a prompt into a production-ready native iOS and Android app in under three minutes.

Vectra AI Delivers Unified Network Observability Across All Major Cloud Platforms to Stop Hybrid Threats Faster

Vectra AI has expanded its platform to provide unified network observability across AWS, Azure, GCP, and OCI, consolidated into a single view. This enhancement helps security teams detect and respond to hybrid cloud threats more rapidly.

Touch Ignites Physical Intelligence: Xense Robotics Showcases the Real Ability of Tactile Intelligence at ICRA 2026

Xense Robotics debuted its TacCap-Gripper tactile data acquisition device and showcased real-world robotic manipulation leveraging both tactile and visual feedback. This demonstration advances embodied AI capabilities for unstructured physical tasks.

With new Majorana 2 quantum chip, Microsoft claims dramatic breakthrough in qubit stability - SiliconANGLE

Microsoft claims its new Majorana 2 quantum chip achieves 1,000 times greater qubit stability than its predecessor, extending qubit lifetimes to at least 20 seconds. This advancement positions the company closer to building a scalable quantum computer by 2029.

Croatia launches Europe's first commercial robotaxi service in Zagreb — European Pulse

Project 3 Mobility has launched Europe's first commercial robotaxi service in Zagreb, Croatia. This marks a significant milestone for autonomous mobility on the continent, even though a human safety operator is still present.

Amazon develops a warehouse robot that workers can speak to | The Verge

Amazon has developed its next-generation Proteus warehouse robot, which can now accept natural language task assignments directly from workers. This eliminates the need for specialized software and expands its operational range across fulfillment centers.

Asana launches AI-powered products to help organizations manage human and agent work - SiliconANGLE

Asana introduced an operating system for human-agent teams, featuring an AI agent named Dash that coordinates work across employees and AI agents using the Work Graph data model. This suite aims to streamline workflow and improve collaboration.

Research with immediate practical relevance

Magenta RealTime 2: Open & Local Live Music Models

Google Magenta has released Magenta RealTime 2 (MRT2), an open-weights live music generation model that boasts 15 times lower latency than its predecessor. It can run on standard MacBook hardware with real-time MIDI and audio control capabilities.

Free AI model powers self-spreading worm in enterprise test network

University of Toronto researchers demonstrated that inexpensive, open-weight AI models can power autonomous network worms capable of exploiting known vulnerabilities at scale. This finding suggests a shift in the cybersecurity threat landscape, moving beyond reliance on expensive zero-day exploits.