The Informed Layer · Newsscan
Daily briefing

OpenAI enters chip race; EU eyes AI security

← All issues

Anthropic / Claude ecosystem

Claude memory now works across Chat and Cowork with user control

Anthropic has unified Claude's memory across Claude Chat and Claude Cowork, allowing memories to be shared between products. Users can see and manage what Claude remembers.

Claude Code sparks developer backlash over AGENTS.md rejection

Anthropic's decision to rely on CLAUDE.md instead of the emerging AGENTS.md standard has triggered developer backlash. The coverage includes Anthropic's official response to community criticism.

Survey: Claude Code overtakes GitHub Copilot as most-used AI coding tool

A new AI coding survey reports Claude Code has become the world's most widely used AI coding tool, with usage doubling to 39% and surpassing GitHub Copilot. It also finds 90% of programmers now use AI agents.

Claude Enterprise-Managed Authorization goes GA for MCP connectors

Anthropic's Enterprise-Managed Authorization is now generally available for Claude MCP connectors. It removes per-user OAuth friction by provisioning access via IdP groups.

Frontier model providers

Google DeepMind scales AI-powered farming tools across APAC and Africa

Google DeepMind is deploying its Agricultural Landscape Understanding and Agricultural Monitoring & Event Detection models across 11 countries. The deployment aims to support sustainable farming and policy decisions.

OpenAI integrates WebMCP into Codex for in-browser agent actions

OpenAI has integrated WebMCP into Codex, providing a standardized bridge for autonomous AI agents to interact with web apps directly. The integration reduces reliance on DOM scraping or visual emulation.

OpenAI launches AI Futures for global governance research

OpenAI has introduced AI Futures, a public commentary platform tied to its Strategic Futures team. The initiative signals a structured approach to AI governance discourse.

Meta to launch AI agent platform Hatch to move beyond ads

Meta is reportedly planning to launch Hatch, a consumer AI agent platform, as it diversifies revenue beyond advertising. The move may include premium model offerings.

Grok 4.6 reaches Google Enterprise Agent Platform

Grok 4.6 is now integrated with Google's Enterprise Agent Platform, expanding enterprise AI tooling options. The integration marks a step in cross-vendor enterprise agent support.

OpenAI unveils Jalapeno AI chip to take on Nvidia

OpenAI has introduced Jalapeno, its first custom AI chip, which Sam Altman says is fast and efficient. The chip aims to reduce reliance on Nvidia for inference workloads.

AI developer tooling & infrastructure

Visual Studio August update brings custom AI agents and multi-branch smarts

Microsoft's Visual Studio August update introduces adjustable model thinking, organization-wide custom agents, and improved Git worktree and submodule support. The update targets more efficient multi-branch development.

X Ads MCP gives AI agents write access to live campaigns

X has introduced a production-capable MCP server that allows AI agents to write to live advertising campaigns. Activation control is explicit to prevent unintended changes.

Cloud & platform providers

Tribe AI has joined Google Cloud's Gemini Enterprise for Legal and Financial Services. The partnership ties Tribe AI's implementation capabilities to Google's industry-focused AI context.

Google Cloud has introduced Gemini Enterprise for Legal, tailored to law firms with enterprise-grade security and private data handling. The offering is available in preview.

AI policy, regulation & governance

EU orders leading AI labs to detail security practices

The European Commission has activated AI Act enforcement by requesting security, safety, and copyright compliance details from leading AI labs. The move signals formal oversight of frontier AI development.

Singapore's Ministry of Law and IPOS have launched a public consultation on whether copyrighted works can be used to train AI. The process aims to balance innovation with creators' rights.

US House receives AI Advertising Disclosure Act

A new federal bill, H.R. 10146, has been introduced in the US House to mandate disclosures for consumer-facing AI tools. The legislation targets commercial influences in AI outputs.

China has formulated nearly 200 key standards for AI sector

China's Ministry of Industry and Information Technology says the country has developed nearly 200 key standards for artificial intelligence. The standards aim to guide sector development.

Smart glasses attract Australian privacy regulator scrutiny

Australia's privacy regulator has highlighted potential gaps in the Privacy Act for wearables such as smart glasses. The OAIC points to the need for new norms and duties to regulate surveillance tech.

Industry & market moves

MotherDuck acquires Tower to consolidate data pipeline foundation

MotherDuck has acquired Tower, the startup already powering its data pipelines, to bring the pipeline runtime in-house. The move aims to control a foundational data infrastructure layer.

Ex-DeepMind founders' robotics startup Generalist hits $3B valuation

Generalist, a robotics startup founded by ex-DeepMind researchers, has raised a $200 million extension at a $3 billion valuation. The funding supports its autonomous robotics foundations and GEN-1.5 capability.

DeepSeek nears pre-IPO funding round paving way for 2027 STAR Market debut

DeepSeek is close to a 50 billion yuan funding round that would value the company at about US$74 billion. The round is positioning DeepSeek for a planned 2027 STAR Market IPO.

Accenture to acquire Dutch SAP partner McCoy for mid-market AI push

Accenture has agreed to acquire McCoy, a Dutch SAP partner, adding over 380 professionals to Accenture Edge. The deal targets mid-market SAP modernization and AI-enabled ERP delivery.

AI product & feature launches

Kingston launches hardcore storage to support AI agent local deployment

Kingston has introduced storage solutions aimed at AI agents running in local deployments. The products address data throughput and privacy for local AI workloads.

Indian lab consortium unveils Astha One humanoid locomotion breakthrough

An Indian lab consortium has unveiled Astha One, a 1.7-meter humanoid with 30 minutes of continuous walking and 20% battery efficiency improvement. It uses IIT Madras torque-controlled actuators and an AI gait algorithm.

Physics-first AI model from Accelerated Understanding processes 5 trillion data points

Accelerated Understanding, founded by researchers once linked to a Bezos project, has introduced a physics-focused AI model. It can process trillions of data points in a single prompt, unlike language-focused models.

Alibaba launches QwenWork International Edition globally

Alibaba has launched QwenWork International Edition, extending its all-in-one workplace AI agent to global markets. The edition includes multi-language support and web/desktop access.

Longbridge integrates investment tools into ChatGPT and Claude

Longbridge has integrated personalized investment tools with live market data into ChatGPT and Claude platforms. Users can link portfolio data to AI assistants for investment workflows.

Okta launches Agent SSO for enterprise AI agents

Okta has launched general availability of Agent SSO, which registers AI agents as identities and governs their access to apps and data. The solution aims to secure agentic enterprise workflows.

Skild AI launches S1 robot foundation model that learns tasks from one video

Skild AI has launched S1, a robot foundation model that can learn 10-minute tasks from a single video without fine-tuning. The model represents a breakthrough in in-context learning for robotics.

Research with immediate practical relevance

AI coder gets Doom running on custom CPU designed by GPT-5.6 Sol

An AI coder has demonstrated Doom running on a custom CPU designed by GPT-5.6 Sol in Turing Complete's sandbox environment. The game viewport is overlaid on a pulsing schematic of the CPU.

Quantization-aware healing shrinks AI models while boosting accuracy

Researchers have introduced quantization-aware healing, a technique that produces a 60B 4-bit model outperforming its full-precision counterpart in accuracy. The smaller model is trained with guidance from the original large model.

Qwen3.8-27B open vision-language model released on Hugging Face

A 27B dense vision-language model, Qwen3.8-27B, is now available open and ungated on Hugging Face. It features a large context window and thinking enabled by default.