Agents advance; AI regulation tightens

Anthropic / Claude ecosystem

Claude adds observability tools for MCP connectors in public beta

Anthropic has launched a public beta of observability tools for Claude's MCP connectors, enabling monitoring, debugging, and performance analysis for third-party connectors. This aims to improve reliability and security of agent-tool chains.

Claude Fable 5 now available on Databricks with Unity AI Gateway governance

Anthropic's Claude Fable 5 is generally available on Databricks, fully governed through Unity AI Gateway. This integration allows Databricks customers to use the Mythos-class model within a managed, enterprise-grade data and AI platform.

Frontier model providers

OpenAI Responses API can now search the web for images

The OpenAI Responses API has been updated to return image results alongside text in its web search capability. This allows developers to unify text and image queries in a single API call, expanding multimodal capabilities.

OpenAI plans ChatGPT super app overhaul with agents and coding tools ahead of IPO

OpenAI is reportedly transforming ChatGPT into a multi-tool super app that includes AI agents and coding capabilities, as it moves toward a public listing. The overhaul reflects a strategic shift from a conversational interface to a broader productivity platform.

Google DeepMind launches Robotics Accelerator for 15 European startups

Google's DeepMind Robotics Accelerator program has enrolled its first cohort of 15 European startups, offering equity-free mentorship and up to $350,000 in cloud credits. The initiative aims to foster physical AI innovation across the continent.

Cohere open-sources North Mini Code, an agentic coding model

Cohere has released North Mini Code, a 30-billion-parameter, open-source coding model under Apache 2.0. It uses a Mixture-of-Experts architecture and runs on a single H100 GPU with a 256k token context window, enabling agentic software engineering tasks.

Google Gemini 3.5 Live Translate enables real-time speech-to-speech translation across 70+ languages

Google has launched Gemini 3.5 Live Translate, its most advanced audio model, capable of continuous, real-time speech-to-speech translation for natural conversation. It supports over 70 languages.

AI developer tooling & infrastructure

VS Code 1.124 ships with smarter autonomous AI agents

Microsoft has released Visual Studio Code 1.124, which enhances autonomous AI agent features and Autopilot improvements. The update allows faster, more autonomous agent workflows directly within the editor.

Relai raises $6.9 million for verifiable continuous learning in AI agents

Relai has launched a continual learning platform and raised $6.9 million to make autonomous AI agents more reliable through live in-loop regression controls. The platform enables agents to learn and verify decisions in real time.

Cloud & platform providers

Azure previews ACI AI Sandboxes for agentic compute

Microsoft has signaled a new category of compute with ACI AI Sandboxes, ultra-fast, isolated environments for autonomous AI agents. This extends Azure Container Instances for agentic workloads, offering secure, on-demand execution.

AI policy, regulation & governance

EU tells AI companies how to make AI-generated content detectable

The European Union has issued guidance instructing AI firms on technical methods to make AI-generated content detectable, likely in preparation for upcoming transparency rules under the AI Act. Details of the guidance remain sparse.

FSB proposes 12 sound practices for responsible AI adoption in finance

The Financial Stability Board has published a consultation report proposing 12 sound practices for governing AI in the financial system. The recommendations aim to balance innovation with resilience against systemic risks.

India pushes for dedicated AI law as IT Act proves insufficient

India's government is considering a standalone AI law, as officials argue the 2000 IT Act cannot adequately address risks from AI and social media intermediaries. Consultations with industry are underway.

Australia prepares for crucial review of AI regulation

The Australian government is set to conduct a significant review of its AI regulation framework, potentially leading to new oversight of AI and data centers. The review aims to position Australia in the global AI race.

Industry & market moves

OpenAI files confidentially for IPO to raise billions

OpenAI has confidentially filed for an initial public offering in the United States, aiming to raise approximately $60 billion at a valuation near $1 trillion. The move follows Anthropic's similar step, signaling a maturing AI industry heading to public markets.

xAI has recruited a Starlink engineer to oversee training operations for its Grok model, signaling a shift toward industrial-scale MLOps and resilience-focused AI development. The move integrates aerospace-system thinking into AI model training.

DeepSeek and Moonshadow push valuations toward $100 billion amid funding rounds

Chinese AI unicorns DeepSeek and Moonshadow are advancing funding negotiations that could push their valuations to the $100 billion level. DeepSeek is attracting interest from national AI industry funds and Tencent.

Warner Music Group acquires Sureel AI to safeguard artist IP

Warner Music Group has acquired Sureel AI, a technology company that helps protect artists' image and work from unauthorized use by AI models. The deal bolsters WMG's capability in IP provenance and monetization in AI contexts.

Poetic emerges from stealth with $50 million from OpenAI for enterprise automation

AI startup Poetic has come out of stealth mode with a $50 million funding round led by OpenAI, targeting enterprise task automation. The company is expected to integrate deeply with OpenAI's ecosystem.

AI product & feature launches

NVIDIA accelerates Google DeepMind’s DiffusionGemma for local AI on RTX GPUs

NVIDIA has optimized Google's DiffusionGemma, a diffusion-based text generation model, to run faster on RTX GPUs and across platforms from local PCs to the cloud. This enables low-latency, private AI inference without requiring data center resources.

Nexla launches MCP Studio for governed, task-specific MCP servers

Nexla has introduced MCP Studio, a solution that dynamically builds governed MCP servers across more than 600 enterprise systems through a single conversational interface. It enables standardized, secure connections for AI agents.

WorkClaw introduces collaborative AI coworkers for Slack and Teams

WorkClaw has launched a platform that brings proactive, customizable AI coworkers into Slack and Microsoft Teams. The virtual assistants can operate autonomously or in concert with human teams, now available in early access.

GENISOM AI reveals full-stack embodied intelligence system at ICRA 2026

GENISOM AI debuted its M1 quadruped robot and a full-stack ecosystem—including actuators, simulation, navigation, and an AI agent framework—at ICRA 2026 for real-world deployment.

NVIDIA announces Cosmos 3, the world's first fully open AI omnimodel

NVIDIA has unveiled Cosmos 3, a fully open omnimodel designed for physical AI applications such as robotics and autonomous vehicles. It combines perception, reasoning, and action in a single framework.

Decart unveils Oasis 3, a photorealistic driving world model

Decart has launched Oasis 3, a real-time, photorealistic world model for driving simulations, with API access for developers and autonomous vehicle simulators. It can generate hours of consistent driving scenes.

McDonald's pilots AI drive-thru chatbot that recognizes regular customers

McDonald's is testing ArchIQ, an AI-powered chatbot at drive-thru lanes that can identify repeat customers and take multilingual orders. The pilot demonstrates a new consumer-facing AI application in fast food.

MotherDuck adds AI-driven natural language data ingestion to cloud analytics

MotherDuck has introduced Flights, an AI-driven workflow tool that allows users to ingest data pipelines via natural language commands within its cloud analytics platform.

Research with immediate practical relevance

DeepSeek expands R1 paper to 86 pages with full training pipeline

DeepSeek has updated its R1 research paper from 22 pages to 86, adding complete training pipeline details, benchmarks, and technical appendices. This comprehensive release aims to facilitate reproducibility and advance open research.

Google introduces DiffusionGemma, a diffusion-based text generation model

Google has open-sourced DiffusionGemma, an experimental model that leverages diffusion processes to speed up text generation by up to 4x on GPUs. It is designed for research and local workflows.

Humanoid robot Pemba scales 20,000-foot volcano, next target Everest

A modified Unitree G1 humanoid robot named Pemba successfully ascended the 20,341-foot Chimborazo volcano in Ecuador, marking the first humanoid to reach such an altitude. The team plans to attempt Mount Everest next.

Waymo creates Reference Driver model to benchmark autonomous driving safety

Waymo has developed the Reference Driver (ReD) model, a virtual human driver benchmark that simulates how a competent human would react to road surprises. The goal is to standardize safety expectations for robotaxis.