The Informed Layer · Newsscan
Daily briefing

Frontier models diversify; Agent security emerges

← All issues

Anthropic / Claude ecosystem

Anthropic Launches Claude Design for Quick Visual Creation and Prototyping

Anthropic introduced Claude Design, an experimental product that enables users to create visuals like prototypes, slides, and one-pagers using natural language prompts within the Claude environment. Powered by Claude Opus 4.7, it's designed to help non-designers rapidly visualize ideas and integrates with tools like Canva for further editing. It also supports applying a team's design system for consistent visual style.

Anthropic CEO to Meet White House Aide Amid Pentagon Dispute Over Mythos AI

Anthropic CEO Dario Amodei is scheduled to meet White House chief of staff Susie Wiles on Friday, signaling a potential breakthrough in the AI startup's dispute with the Pentagon. The meeting comes as the Trump administration acknowledges the advanced cybersecurity capabilities of Anthropic’s Mythos AI model for defense, despite a previous ban on the Pentagon working with Anthropic.

Frontier model providers

OpenAI Launches GPT-Rosalind for Life Sciences Research and Drug Discovery

OpenAI has introduced GPT-Rosalind, a new reasoning model specifically designed for biology, drug discovery, and translational medicine. Available as a research preview to qualified enterprise customers in the U.S., it aims to accelerate scientific workflows from hypothesis generation to experimental planning. It also includes a free Life Sciences research plugin for Codex, offering access to over 50 data sources and tools.

Google Gemini Now Generates AI Images from Personal Data Using Google Photos

Google's Gemini app is rolling out a 'Personal Intelligence' feature that enables its Nano Banana 2 image generation model to create personalized AI images by drawing on users' Google Photos libraries and connected Google apps. This eliminates the need for detailed prompts or manual uploads, allowing Gemini to generate images reflecting a user's taste and lifestyle, with privacy features ensuring photos are not used for model training.

Mozilla Rolls Out Thunderbolt Enterprise AI Client for Sovereign AI Deployments

Mozilla subsidiary MZLA Technologies has released Thunderbolt, an open-source enterprise AI client designed to allow organizations to keep AI models, internal systems, and automation tools within their own infrastructure. Thunderbolt offers a unified interface for chat, search, and research, supports open protocols like Model Context Protocol (MCP) and Agent Client Protocol (ACP), and runs as native apps across various operating systems.

Meta Launches Muse Spark AI with Image Understanding and Parallel Task Handling

Meta introduced Muse Spark, its foundational AI model, now powering the Meta AI assistant across Meta's apps and AI glasses. The model features new Instant and Thinking modes for varying task complexity, the ability to launch multiple subagents in parallel for comprehensive responses, and strong multimodal perception for understanding images. It also includes dedicated shopping and enhanced health Q&A modes.

AI developer tooling & infrastructure

Cloudflare Unveils AI Platform as Unified Inference Layer with Multi-Provider Access

Cloudflare has transformed its AI Gateway and Workers AI into a unified inference layer, allowing developers to access over 70 AI models from more than 12 providers (including OpenAI, Anthropic, Google, Alibaba Cloud, Bytedance) through a single API. This platform provides automatic failover, cost monitoring across providers, and features for faster time-to-first-token in agent workflows, along with private beta access to bring-your-own-model capabilities.

Cloudflare Launches Artifacts: A Git-Compatible, Versioned Storage for AI Agents in Beta

Cloudflare introduced 'Artifacts,' a distributed, versioned filesystem that speaks Git, designed for AI agents. It allows programmatic creation of repositories for each agent session or sandbox instance, supporting cloning, forking, and committing through both Git clients and a REST API. Built on Durable Objects and WebAssembly, Artifacts aims to persist agent state, manage large repos efficiently, and enable collaboration with features like git-notes for metadata.

Cloudflare Email Service Enters Public Beta, Enabling Email-Native AI Agents

Cloudflare's Email Service is now in public beta, providing full bidirectional email capabilities for applications and AI agents. It allows sending transactional emails directly from Workers or via a REST API, with automatic configuration of SPF, DKIM, and DMARC. Combined with the Agents SDK, it enables email-native agents to receive, process, and reply to emails asynchronously, supported by new Email MCP server, Wrangler CLI commands, and an open-source Agentic Inbox reference app.

Microsoft Introduces Hybrid AI Automation in Copilot Studio for Enhanced Enterprise Workflows

Microsoft has launched new capabilities in Copilot Studio that combine AI agents and traditional workflows to enhance automated operations. This hybrid approach allows workflows to call agents for judgment calls on structured automations via 'agent nodes,' and conversely, enables agents to use workflows as tools for executing subprocesses within complex tasks. This addresses the limitations of pure AI autonomy in production environments.

Prefactor Launches AI Agent Management Platform for Governance, Security, and Auditing

Prefactor has emerged from early access with an AI Agent Management Platform, designed as an 'agent runtime control plane.' It offers capabilities to track, assess, and act on AI agent activity across various frameworks (e.g., Claude, OpenAI, LangChain, CrewAI), providing real-time identity, policy enforcement, and full audit trails. The platform aims to address visibility gaps, control vacuums, and trust issues as enterprises deploy AI agents in production.

Microsoft Fabric Announces General Availability of OneLake MCP for AI Agent Data Access

Microsoft Fabric has made its OneLake Model Context Protocol (MCP) tools generally available, allowing AI agents to interact with all data in OneLake (from lakehouses to mirrored databases) through natural language conversations. The tools provide file system APIs, table APIs, and workspace/item discovery, operating under existing Azure identity and Fabric permissions. This enables agents to document schemas, map storage, and assess database health without code or clicks.

Capsule Security Raises $7M Seed Round to Guard AI Agents Against Runtime Risks

Capsule Security emerged from stealth with a $7 million seed round to fund its security product designed to monitor and control AI agents in real-time. The platform aims to enforce trust at runtime, evaluating actions in context and blocking unsafe or unauthorized activities before completion. Capsule also disclosed vulnerabilities (ShareLeak and PipeLeak) in Microsoft Copilot Studio and Salesforce Agentforce, and released ClawGuard, an open-source enforcer for OpenClaw.

Cloud & platform providers

Databricks and Google Cloud Showcase Gemini Integrations and AI-Driven Data Platform

Databricks and Google Cloud highlighted their partnership at Google Cloud Next 2026, focusing on their unified platform for AI-driven business. Databricks has become a first-party provider of Gemini models, offering direct API access within the Databricks platform for building, customizing, and deploying agentic AI on governed enterprise data. The collaboration emphasizes accelerating AI adoption beyond proof-of-concept into production, with examples of cost optimization and regional compliance.

Crusoe Announces 900 MW Abilene AI Factory Campus for Microsoft

Crusoe, a neocloud provider, revealed plans for a new 900 MW AI factory campus in Abilene, Texas, dedicated to supporting large-scale AI workloads for Microsoft. The campus, which will eventually reach 2.1 GW, includes two new buildings and an on-site power plant for grid resilience and energy-efficient infrastructure. The first building is expected to be operational by mid-2027.

Google Kubernetes Engine (GKE) Issues New Security Bulletins for Linux Kernel Vulnerabilities

Google Kubernetes Engine (GKE) has issued multiple new security bulletins (GCP-2026-020, GCP-2026-010, and updates to several others dating back to 2021) addressing a range of high-severity vulnerabilities in the Linux kernel. These flaws, including CVE-2026-23231 and CVE-2023-6931, could lead to privilege escalation on Container-Optimized OS and Ubuntu nodes across GKE, GKE on VMware, GKE on AWS, and GKE on Azure. Patch versions and upgrade instructions are provided.

Microsoft Outlines Azure Cloud Strategy to Win Long-Term AI Deals with Flexibility

Microsoft is aiming to replicate its Azure cloud business strategy of flexibility to secure long-term AI deals with clients. Jay Parikh, EVP of Microsoft’s global leadership team, stated that the go-to-market approach is to offer a fully flexible AI platform, allowing clients to choose between the safest AI models and the most cutting-edge 'frontier' AI models. This strategy is based on building strong client partnerships.

FOX Corporation Names AWS as Preferred AI Cloud Provider, Deepening Strategic Collaboration

FOX Corporation has announced a strategic collaboration with Amazon Web Services (AWS), designating AWS as its preferred AI Cloud provider. This partnership will power FOX's direct-to-consumer platform, FOX One, by integrating AWS Elemental MediaLive, MediaPackage, and MediaTailor with AWS AI/ML services, including AWS Elemental Inference. The aim is to drive AI-driven innovations such as content enrichment, highlight generation, and personalized discovery across FOX's media portfolio.

Schneider Electric and Microsoft Unveil Azure AI Factory Tools for Industrial Automation

Schneider Electric has partnered with Microsoft to launch new manufacturing tools developed with Azure AI, aimed at industrial engineering and operations. The tools integrate with Schneider Electric's EcoStruxure Automation Expert platform, using specialized AI agents and an industrial copilot built with Azure AI to automate routine design choices, validate automation logic, and reduce time in control configuration by up to 50%.

AI policy, regulation & governance

Australian Court Warns Lawyers on AI Use, Emphasizing Accuracy and Disclosure

The Federal Court of Australia has issued new guidelines warning lawyers about the careless use of AI in legal work, citing a rise in errors like fake citations and incorrect content in AI-generated material. The court emphasizes lawyers' responsibility for accuracy, mandating double-checking AI outputs and upfront disclosure of AI assistance, also raising privacy concerns about feeding sensitive data into AI systems.

India Establishes AI Governance and Economic Group (AIGEG) to Align National Policy

India's Ministry of Electronics and Information Technology (MeitY) has formed the AI Governance and Economic Group (AIGEG), a high-level inter-ministerial body chaired by Union Minister Ashwini Vaishnaw. This apex body will coordinate AI policymaking across ministries, review regulatory frameworks, issue guidelines for compliance, and develop a roadmap for AI deployment, including assessing job impacts and workforce transitions.

EU AI Office Lacks Access to Anthropic's Mythos Model, UK AISI Leads in Evaluation

The European Commission's AI Office has confirmed it does not have access to Anthropic's powerful Mythos AI model, unlike the UK's AI Security Institute (AISI), which published a technical risk analysis within a week. This highlights a significant gap in the EU's ability to evaluate frontier AI risks, with critics arguing the AI Office is understaffed and too far from political leadership to respond effectively to such advanced model releases.

US Lawmakers Express Existential Fears About AI in Congressional Roundtable

A U.S. congressional subcommittee held a roundtable discussion on AI where lawmakers expressed significant anxieties, ranging from federal workers using AI chatbots for sensitive data and AI-generated pornographic images, to AI systems denying lethal actions in the military and its intensive energy usage. The discussion included fears of AI leading to 'destruction' and a 'revolution' if not properly regulated.

Industry Calls for Simpler AI Rules and Clearer Timelines in Europe

Fifteen trade associations, including BSA and AmCham EU, have issued a joint statement urging Brussels for a swift agreement on the Digital Omnibus on AI, seeking to streamline the bureaucratic burden on businesses. They advocate for extending the grace period for labeling generative AI under Article 50 to 12 months and exempting non-high-risk systems from EU database registration to avoid stifling innovation.

European Commission Investigates Meta Over WhatsApp AI Assistant Access

The European Commission has sent Meta a Supplementary Statement of Objections, signaling its intent to order the company to restore third-party AI assistants' access to WhatsApp under terms prior to Meta's October 2025 policy change. The Commission argues Meta's conduct forecloses competition in the AI assistant market, despite Meta having reversed its original exclusion and allowing third-party AI assistants under paid access terms.

Bank of England Plans AI-Agent Stress Tests After MP Warnings on Market Instability

The Bank of England has committed to investigating the impact of AI agents demonstrating correlated or 'herding' behavior in financial markets, directly responding to concerns from the Treasury Committee. The FCA will also share AI best-practice examples with financial services firms. This follows criticism regarding the clarity of current guidance and Treasury's delay in bringing AI providers into the Critical Third Parties Regime.

Industry & market moves

Sequoia Raises $7 Billion Fund to Double Down on Late-Stage AI Bets

Sequoia Capital has closed a $7 billion fund, its largest for late-stage investments in the US and Europe, under new co-stewards Alfred Lin and Pat Grady. This significant capital raise aims to fund investments in AI infrastructure and applications, including robotics startups, reflecting a broader shift in venture capital towards mega-rounds for foundation models and AI-driven companies.

Hyperscale Data Partners with AGIBOT for AI Robotics and Data Collection

Hyperscale Data (NYSE: GPUS), through its subsidiary Omnipresent Robotics, announced a strategic partnership with AGIBOT, a developer of intelligent robotics technology. The collaboration will focus on deploying and commercializing intelligent robotic systems and AI data collection initiatives, with further details expected to be released on April 20–21, 2026.

K1x Secures $175 Million Growth Investment to Scale AI-Native Private Market Tax Compliance Platform

K1x, an AI-native private markets tax data platform, closed a $175 million growth investment led by Sumeru Equity Partners, with continued backing from Edison Partners. This funding will be used to deepen core platform capabilities, accelerate product development in machine learning and generative AI, and strengthen customer support. K1x's platform automates the extraction, aggregation, and standardization of complex K-1 data for private markets, addressing an estimated $27 billion annual burden.

Nvidia AI Chip Rivals Attract Record Funding Amid Focus on Inference

Startups challenging Nvidia's dominance in AI chips raised a global record of $8.3 billion in 2026, driven by a shift in focus towards efficient AI inference (deployment in applications). Companies like Cerebras Systems, MatX, Ayar Labs, and Etched secured large funding rounds, with European rivals Euclyd and Optalysys also planning significant raises. This comes as Nvidia itself invests heavily in AI inference technologies and photonics.

Cathay Financial Holdings Partners with OpenAI for Enterprise-Wide AI Adoption

Cathay Financial Holdings (Cathay FHC) has announced a long-term strategic collaboration with OpenAI, becoming the first financial institution in Taiwan to integrate advanced AI at this scale. The partnership involves deploying ChatGPT Enterprise across its subsidiaries and co-developing agentic AI systems for corporate banking and insurance, with a focus on secure, enterprise-grade AI tools and strengthening employees' AI proficiency.

American Express Acquires AI Expense Management Startup Hyper

American Express is reportedly acquiring Hyper, an AI-powered expense management startup. While specific terms were not disclosed, Hyper's product offered AI-driven automation for expense reporting and financial tracking.

Ori Industries Merges with Radiant to Form AI Infrastructure Platform with Brookfield Funding

Ori Industries, a Data Sales portfolio client, has merged with Radiant, a Brookfield portfolio company, to create a new AI infrastructure platform. The new entity is purpose-built for large-scale artificial intelligence workloads and will have access to substantial institutional funding from Brookfield, along with integrated data center and energy development capabilities, to build and scale AI infrastructure at a lower cost.

Research with immediate practical relevance

AI Helps Instructors Give Better Feedback, Improving Student Revisions in Trial

A randomized trial in a large economics course at the University of Michigan found that an AI-mediated system called FeedbackWriter significantly improved student revisions. This system offers teaching assistants (TAs) AI suggestions aligned with assignment rubrics, with TAs retaining final control. The study suggests AI can help instructors provide higher-quality, more systematic feedback, leading to better learning outcomes, rather than replacing human graders.