The Informed Layer · Newsscan
Daily briefing

Regulation expands; New AI agents enter testing

← All issues

Anthropic / Claude ecosystem

Anthropic ships major Claude Design overhaul with design system imports, code round-trips, and a fix for its token-burning problem

Anthropic has released a significant overhaul of Claude Design, transforming it into an enterprise-grade brand-compliance tool. The update features bidirectional integration with Claude Code, allowing seamless design-to-engineering handoff, and addresses previous issues like token burning, improving efficiency and usability.

Frontier model providers

A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry

OpenAI's GPT-5.4 has demonstrated the ability to autonomously identify TEMPO as an additive, improving the yield of Chan-Lam coupling for primary sulfonamides from 16.6% to 25.2% mean yield. This discovery was subsequently validated through bench-scale experiments, highlighting the model's potential for scientific research.

AI developer tooling & infrastructure

WebMCP enters public testing in Chrome 149

Google has advanced WebMCP (Web Model Context Protocol), an open web standard for AI agents, from prototype to public testing. It is now available via Chrome 149's origin trial, allowing developers to expose structured tools to AI agents for the first time.

Vercel Releases Eve: An Open-Source AI Agent Framework Where Each Agent is a Directory of Files Mapped to Capabilities

Vercel has launched Eve, an open-source AI agent framework that organizes agents as directories of files, with capabilities mapped directly to their structure. Eve includes built-in features for durable execution, sandboxing, and observability.

Cloud & platform providers

Introducing Amazon Bedrock Managed Knowledge Base for faster, more accurate enterprise AI applications

AWS has launched Amazon Bedrock Managed Knowledge Base, abstracting the complexities of RAG (Retrieval Augmented Generation) pipelines into a single managed service. This offering includes native data connectors, smart parsing, and agentic retrieval capabilities, designed for faster and more accurate enterprise AI applications.

Announcing General Availability of the Azure Cosmos DB Built-in Connector for Logic Apps Standard

The Azure Cosmos DB built-in connector for Logic Apps Standard is now generally available. This connector provides native change feed triggers, bulk operations, and Entra ID authentication, enhancing enterprise integration workflows.

Workers tracing now supports custom spans · Changelog

Cloudflare Workers now offers support for custom trace spans through its new tracing.enterSpan() API. This feature allows developers to instrument application-specific logic, integrating it with the platform's automatic instrumentation for more granular observability.

AI policy, regulation & governance

Anthropic’s new privacy policy offers US consumers a way around the Fable ban

Anthropic has updated its privacy policy to include a provision for identity verification via government ID. This change could potentially allow US consumers, who were previously restricted, to regain access to Anthropic's Fable and Mythos models, which are currently under a US export ban for foreign nationals.

Top AI Executives Gather in France Amid Calls for Tech Sovereignty in Europe

Leading AI executives from OpenAI, Google DeepMind, and Anthropic convened at the G7 summit in France to discuss tech sovereignty and the need for a European alternative to dominant US AI platforms. This comes amid growing geopolitical tensions and calls for greater control over critical AI infrastructure.

Justice Department seeks to dismiss lawsuit against xAI power plant, citing national security

The Department of Justice has invoked national security to defend xAI's unpermitted gas turbines, which power classified military AI operations. This action sets a precedent where national security concerns related to AI computing may override environmental enforcement, in response to a Clean Air Act lawsuit (NAACP v. xAI Colossus 2).

Trump Administration Delays Blacklisting Over 100 Chinese Firms, Including DeepSeek

The Trump administration has delayed the public announcement of new Entity List additions, which were set to target over 100 Chinese companies, including AI giant DeepSeek. This marks the longest hiatus in such designations in over a decade, reportedly to avoid escalating trade tensions with China.

Department of the Navy announces cloud computing task orders for Joint Warfighting Cloud Capability (JWCC) contract

The Department of the Navy has awarded cloud computing task orders to multiple providers including AWS, Azure, GCP, and Oracle under the Joint Warfighting Cloud Capability (JWCC) contract. These orders aim to streamline mission-critical cloud access and accelerate digital transformation efforts.

Digital Omnibus: EU Parliament Delays the AI Act Rules

The EU Parliament has approved the Digital Omnibus, a package of amendments to the AI Act, which will delay high-risk AI Act compliance deadlines until 2027–2028. The omnibus also introduces new prohibitions concerning non-consensual intimate imagery and CSAM.

Why we have six months to regulate the AI that Europe isn’t building – CEPS

A CEPS analysis argues that Europe's AI Act currently lacks the enforcement mechanisms necessary to regulate open-source models developed by non-EU entities. This creates a critical six-month window before Chinese open-source models potentially close the capability gap with Western counterparts, posing significant regulatory challenges.

India Tightens Oversight On AI-Generated Content Under IT Rules

India's Ministry of Electronics and Information Technology (MeitY) has proposed a draft amendment to the IT Rules, 2021, to tighten oversight on AI-generated content. The amendment mandates continuous and clearly visible labeling for all synthetic information across intermediary platforms, including deepfakes, voice cloning, and face-swap modifications.

G7 leaders vow closer ties on AI

G7 leaders have committed to closer coordination on AI and proposed a 'trusted partners' scheme. This initiative aims to allow non-US allies access to advanced US AI models, despite existing Trump administration restrictions on foreign access, fostering international collaboration in AI development.

Pentagon admits using Elon Musk’s chatbot Grok to launch strikes on Iran

The US Pentagon has officially admitted to utilizing xAI's Grok chatbot (specifically the Grok Gov Model) to coordinate and launch over 2,000 munitions strikes against Iranian targets within a 96-hour period.

Industry & market moves

HSBC AND GOOGLE CLOUD ANNOUNCE TRANSFORMATIVE AI BANKING PARTNERSHIP

HSBC and Google Cloud have formed a multi-year AI partnership with the goal of deploying over 200 AI use cases within two years. The collaboration targets more than $100 million in value from high-priority initiatives across wealth management, financial crime detection, and frontline operations.

Intelligence artificielle - Québec s'associe avec Mistral pour explorer le potentiel de l'IA dans la transformation numérique de l'État

The Quebec government has entered into an exploratory partnership with Mistral AI. This collaboration aims to advance digital sovereignty and integrate AI responsibly into public administration, building on a similar agreement previously made with Cohere.

DeepSeek Raises $7.4 Billion, Only the Chinese State Gets Voting Rights

DeepSeek has successfully closed its first external funding round, raising $7.4 billion at a valuation that makes it China's most valuable AI startup. A notable condition of this Series A funding is that the Chinese state retains exclusive voting rights, ensuring founder control while securing significant capital.

ELCA collaborates with Microsoft to Expand its Sovereign Cloud Options in Switzerland with Locally Operated Services Built on Microsoft Azure Local and Microsoft 365 Local

ELCA is expanding its sovereign cloud offerings in Switzerland by locally operating Microsoft Azure Local and Microsoft 365 Local services. This collaboration with Microsoft aims to meet strict Swiss data sovereignty and compliance requirements while facilitating AI adoption for local enterprises.

Odyssey raises $310M at $1.45B valuation to transform AI model simulation

Odyssey has successfully raised $310 million in Series B funding, achieving a $1.45 billion valuation. The funding will be used to advance world models, a new category of foundation AI systems capable of simulating physics and causality for applications in robotics and autonomous systems.

Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it.

XDOF has emerged from stealth with $70 million in Series B funding to build infrastructure for collecting robot training data. This development comes as AI labs intensify their efforts to develop physical robotics capabilities, recognizing the critical need for high-quality, real-world data.

Sharon AI Announces Oversubscribed US$1.6 Billion Strategic Financing to Accelerate Expansion of AI Factories Across Australia and the Asia-Pacific

Australian AI infrastructure company Sharon AI has secured an oversubscribed US$1.6 billion private placement financing. This funding will be used to deploy over 55,000 NVIDIA GPUs across AI factories in Australia and the Asia-Pacific region, significantly expanding its computing capacity.

Backed by Bezos and AI godfather, Yann LeCun – the little known Cambridge AI start up chasing a $400m raise

CuspAI, a Cambridge-based materials-discovery AI startup backed by Jeff Bezos and AI pioneer Yann LeCun, is seeking to raise $400 million in Series B funding. The company aims to address the physical infrastructure bottlenecks impacting AI chip, cooling, and water resource constraints in hyperscale data center expansion.

Barcelona-based NeuralTrust raises €17.2 million to secure and govern enterprise AI agents

NeuralTrust, a Barcelona-based AI security startup, has closed a €17.2 million seed round, making it the largest cybersecurity seed financing by an EU company. The funding will be used to develop solutions for securing and governing enterprise AI agents.

Investing in Convey | Andreessen Horowitz

Andreessen Horowitz has led a $38 million Series A funding round for Convey, an AI agent platform that has already accumulated 1.1 million hours of real work within enterprise customers like NBCUniversal and Unity.

Nano Dimension Signs Term Sheet with Infinite Epigenetics to Form a Publicly Traded, AI-Powered Preventive Health and Diagnostics Company

Nasdaq-listed Nano Dimension has signed a non-binding term sheet to acquire Infinite Epigenetics, an AI-powered epigenetics diagnostics platform. The proposed merger aims to create a publicly traded healthcare AI company focused on preventive health and diagnostics.

AI product & feature launches

Unlocking UK house-building with AI-accelerated planning

Google DeepMind, in collaboration with the UK government, is deploying AI-accelerated planning tools. This initiative aims to halve the processing time for planning applications, thereby accelerating house-building projects across England and addressing housing shortages.

Continuous AI Pentesting: What We’re Building, and What It’s Already Finding

CyCognito's continuous AI pentesting platform has identified real-world exposures, including misconfigured MCP servers, vulnerable RAG indices, and weaknesses in physical security systems. The platform achieves this by automating multi-step attack chains at scale.

What’s new in Databricks Platform security and compliance at Data + AI Summit 2026

Databricks has introduced Automatic Identity Management (AIM) for Entra ID, now generally available on AWS and GCP, and Context-Based Ingress policies. These new features are designed to simplify enterprise-scale security and compliance across diverse data and AI workloads.

Spur Enhances Monocle Session Enrichment with No-Code Cloudflare Integration to Accelerate Inline Traffic Enforcement

Spur has introduced a no-code Cloudflare integration for its Monocle Session Enrichment platform. This integration enables real-time inline traffic enforcement for anonymized traffic detection without requiring extensive engineering resources.

MATSUMI, a Holographic AI Presence for XR Glasses, Announced at AWE 2026

MATSUKO has announced MATSUMI, a holographic AI companion for XR glasses, at AWE 2026. This marks the first AI companion rendered as a photorealistic holographic spatial presence, rather than a chat interface, enabling embodied interaction across major XR platforms like Apple Vision Pro and Meta Quest.

Nvidia's AI squadmate is finally dropping into PUBG

Nvidia ACE (Avatar Cloud Engine), an NPC system powered by a small language model, is now available for real-time squad co-op gameplay in PUBG. This integration brings dynamic speech synthesis and interactive behavior to in-game companions.

Mitiga Labs Launches Skillgate to Identify Risks in AI Agent Skills and Configurations

Mitiga Labs has launched Skillgate, a free scanner designed to detect security risks in AI agent skills and configurations. It identifies prompt injection, remote code execution (RCE), credential exfiltration, and other attack techniques before agents are deployed.

Research with immediate practical relevance

MolmoMotion: Language-guided 3D motion forecasting | Ai2

The Allen Institute for AI (AI2) has released MolmoMotion, a language-guided 3D motion forecasting model. It is trained on the largest corpus of action-described object trajectories (1.16 million videos) and features a human-validated benchmark for object-centric motion prediction.

SandboxAQ, Led by CEO Jack Hidary, Publishes Spin-Aware Catalysis AI Research

SandboxAQ has published research on spin-aware catalysis AI, featuring the AQCat25 dataset and model. This is the first large-scale catalysis dataset to incorporate spin polarization for earth-abundant magnetic metals, enabling physics-based AI modeling that is 20,000 times faster than first-principles simulation.

This AI Learned the Laws of Physics and Could Accelerate Quantum Computing Breakthroughs

Researchers at Chalmers University of Technology have developed an AI that learns and embeds the laws of physics directly into neural networks for optical component design. This method reduces training time by 90% and is expected to accelerate breakthroughs in quantum computing applications.