Agents drive infra, funding, and regulation; OpenAI ships…

Anthropic / Claude ecosystem

Claude Tag Introduces Agent Identity for Autonomous Team AI - Blockchain.News

Anthropic has launched 'agent identity' for Claude Tag, an access model enabling secure autonomous AI collaboration in team workspaces. This new feature provides workspace-level permissions, removing the reliance on individual user credentials for AI agent access.

Anthropic's Cherny: recursive agent loops are the next coding era | AIntelligenceHub

Anthropic's Claude Code leader, Cherny, posits that recursive agent loops, where AI agents continuously prompt and interact with each other, will define the next major paradigm shift in software development in 2026. This approach aims to accelerate how software teams ship code.

Frontier model providers

No significant new developments.

AI developer tooling & infrastructure

Introducing New Relic Preflight for AI Coding Observability

New Relic has launched Preflight, an open-source Model Context Protocol (MCP) server designed to provide observability for AI coding assistants. Preflight offers token-cost visibility, anti-pattern detection, and efficiency scoring for tools like Claude Code, Cursor, and GitHub Copilot.

Cursor Trains First Frontier Model From Scratch on Colossus: 1.5 Trillion Parameters

Cursor, previously a reseller of API models, has trained its first frontier AI model from scratch using xAI's Colossus supercomputer. The new model boasts 1.5 trillion parameters, marking a significant transition to proprietary foundation model ownership.

Infragistics Puts MCP Toolchain at Center of Ultimate 26.1 -- Visual Studio Magazine

Infragistics has integrated Model Context Protocol (MCP) servers into its Ignite UI Ultimate 26.1, centralizing a toolchain that provides AI coding assistants with rich context from component libraries, documentation, and theming systems. This supports Angular, React, Web Components, and Blazor development.

Coinbase reduces time from idea to production by 90% with Cursor · Cursor

Coinbase reports a 90% reduction in engineering cycle time, from 20 days to under 2 days, by implementing Cursor's agentic infrastructure across its 2,400-developer team. The company redesigned workflows to delegate implementation to AI agents, allowing engineers to focus on intent and validation.

Oracle Launches Free Agentic AI Foundations Certification for Developers, TechGig

Oracle University has introduced a free certification program, 'Oracle Agentic AI Foundations,' aimed at equipping developers with the necessary skills to build and deploy AI agents. The curriculum covers frameworks like LangChain and OpenAI Agent stack, integrated with OCI Enterprise AI.

Major Product Release v0.91.0 | What's New | OpenObserve

OpenObserve v0.91.0 has been released, introducing AI agent observability with support for the Model Context Protocol (MCP) and a redesigned user interface across the entire platform.

Prime-rl 0.6.0 Brings Trillion-Scale Agentic RL - Cosmic Meta Digital

Prime Intellect's prime-rl 0.6.0 now enables trillion-parameter Mixture-of-Experts (MoE) model training in agentic Reinforcement Learning (RL) workloads. Using 28 H200 nodes, it achieves sub-five-minute training steps, moving large-scale RL from theoretical to practical application.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

France Advances AI Ambitions With 28 New Projects; Asks Mistral AI CEO Whether It’s Too Late to Matter - The Paris Fool

France has announced 28 new AI projects under its 'France 2030 Pionniers de l'IA' national strategy. These initiatives aim to establish France as a significant AI hub by developing specialized applications in healthcare, agriculture, and legal sectors, rather than directly competing in frontier model development.

The US is pressing Meta to let it review its AI, and Meta is the last holdout

Meta is currently the last major US AI developer to resist submitting its models for government security review, following a new Trump executive order. This creates a standoff as the administration signals potential enforcement through export controls.

Legal tech firm Legion LegalTech is challenging the US government's directive restricting foreign nationals' access to Anthropic's Fable 5 and Mythos 5 models. The firm argues that export-control laws should not apply to hosted AI models.

Five Eyes Just Issued an AI Warning. The EU Already Made It Law. Where Does That Leave Australia & New Zealand?

The EU AI Act's high-risk compliance requirements are now fully active, imposing binding obligations that can exceed GDPR penalties and apply extraterritorially to providers serving EU markets. This contrasts sharply with Australia and New Zealand's current voluntary AI governance frameworks.

AI Governance for Law Firms: What VLSB+C, LPLC and OAIC Require

Three Australian legal regulators—VLSB+C, LPLC, and OAIC—have outlined their expectations for AI governance within law firms. These expectations are communicated through existing professional conduct, insurance, and privacy frameworks.

Privacy Commissioner finds privacy breaches in third-party tracking pixel investigation | OAIC

The Office of the Australian Information Commissioner (OAIC) has ruled that third-party tracking pixels collecting sensitive health information without consent violate the Australian Privacy Act. This establishes a significant precedent for health service websites.

Industry & market moves

OpenAI and Broadcom unveil LLM-optimized inference chip | OpenAI

OpenAI has unveiled 'Jalapeño,' a custom LLM-optimized inference accelerator developed in collaboration with Broadcom. This chip, designed in nine months and set for gigawatt-scale deployment in 2026, promises significantly improved performance per watt compared to current state-of-the-art solutions.

Exclusive: Taktile raises $110 million from Goldman Sachs, Tiger Global to automate high-stakes financial decisions | Fortune

AI startup Taktile has secured $110 million in Series funding from Goldman Sachs and Tiger Global. The investment will fuel Taktile's mission to automate high-stakes financial decision-making processes in the banking and insurance sectors.

EXL to acquire iMerit, advancing its leadership in

EXL (ExlService Holdings, Inc.) is set to acquire AI data company iMerit for up to $310 million. This acquisition will enhance EXL's foundation model training expertise, positioning the company at the forefront of enterprise AI development.

MoEngage Acquires US-Based AI Startup Aampe To Scale Agentic Marketing

MoEngage has acquired US-based AI startup Aampe in an all-cash deal worth tens of millions of dollars. The acquisition aims to scale MoEngage's agentic decisioning capabilities for personalized customer marketing, enabling 1:1 personalization at an individual-agent level.

AI product & feature launches

Tencent DeepSeek Integration: AI Agents in WeCom — The AI Chronicle

Tencent is reportedly integrating DeepSeek models into its WeCom enterprise communication platform to automate various workflows. This move is aimed at reducing reliance on Western AI solutions and enhancing China's technological autonomy amidst ongoing US chip restrictions.

Research with immediate practical relevance

Google DeepMind's Alex Imas says GLM 5.2 uses distillation solely to cold-start reinforcement learning on coding tasks

Google DeepMind's Alex Imas clarified that GLM 5.2's high performance in coding tasks is primarily due to reinforcement learning (RL) hill-climbing on distilled trajectories, rather than the distillation process itself. This technical explanation corrects misconceptions about the model's training methodology.