The Informed Layer · Newsscan
Daily briefing

Agents drive product, infra, policy; funding for sovereign…

← All issues

Anthropic / Claude ecosystem

Anthropic releases Project Fetch Phase Two results: Claude Opus 4.7 completes robotics tasks 10-37x faster

Anthropic's Project Fetch Phase Two demonstrates that Claude Opus 4.7 can complete complex robotic manipulation tasks 10 to 37 times faster than human teams. This significant speed increase showcases rapid advancements in agentic physical AI capabilities, pushing the boundaries of AI-driven robotics.

Anthropic extends Artifacts feature to Claude Code, enabling shared live coding sessions with automatic updates

Anthropic has launched Artifacts for Claude Code, allowing development teams to share interactive web pages generated directly from live coding sessions. This feature provides automatic updates, real-time artifact versioning, and access controls, streamlining collaborative coding workflows.

Frontier model providers

OpenAI launches scheduled tasks in ChatGPT for automated reminders, recurring work, and monitoring

OpenAI has introduced a new scheduled tasks feature in ChatGPT, enabling users to automate reminders, recurring work, and monitoring activities. This update includes a dedicated 'Scheduled' page for easy management and is available for Plus, Pro, Business, and Enterprise users.

OpenAI Academy launches three free AI courses to lower entry barriers for workforce upskilling

OpenAI Academy has launched three free, certificate-bearing AI courses: AI Foundations, Applied AI Foundations, and Agents and Workflows. These courses are designed to lower entry barriers for individuals seeking to upskill in large language model tools and agentic workflows.

OpenAI Unveils Workspace Agents for Continuous Office Operations

OpenAI has launched Workspace Agents, designed to enable continuous autonomous office operations. These agents are intended to streamline and automate various office tasks, offering a new level of efficiency and self-sufficiency in business environments.

Google DeepMind unveils RT-3, a next-generation robotic foundation model for humanoids

Google DeepMind has unveiled RT-3, its next-generation robotic foundation model, which features improved generalization, dexterity, and natural language understanding. Designed for humanoid robots, RT-3 aims to enhance their ability to perform tasks across diverse hardware configurations.

Mistral AI launches hosted AI cloud services with European sovereignty positioning

Mistral AI has officially launched its hosted AI cloud services, positioning itself on European sovereignty. This offering targets enterprises and developers, providing an alternative to U.S.-based hyperscalers with a focus on data governance and localized infrastructure.

xAI's Grok Imagine Video 1.5 accelerates 720p generation to 25 seconds with integrated audio synthesis

xAI has released Grok Imagine Video 1.5, which significantly accelerates 720p video generation to just 25 seconds. The update also includes integrated audio synthesis and new creator workflow tools, positioning it as a speed-focused competitor in the AI video market.

DeepSeek launches web version of its image recognition mode

DeepSeek has officially launched a web version of its image recognition mode, making its visual AI capabilities more accessible. Early testing, however, indicates some challenges in accurately identifying specific individuals.

China's Zhipu AI releases GLM-5.2, outperforming Google's models on open benchmarks at 10x lower cost

China's Zhipu AI (Z.ai) has released GLM-5.2, which ranks as the most capable openly available language model on independent benchmarks. It undercuts U.S. flagship models by 10x on pricing while achieving near-frontier performance in coding and reasoning tasks.

AI developer tooling & infrastructure

ECC v2.0.0 standardizes agent configuration across seven code harnesses, adds unified security scanner

The ECC project has released v2.0.0, a cross-harness agent system that standardizes agent configuration across seven popular AI coding tools including Claude Code, Cursor, and Codex. It also introduces a unified security scanner and a cross-harness parity system to ensure consistent behavior and security.

Cursor launches Origin, a Git forge purpose-built for parallel AI agents

Cursor has launched Origin, a new Git forge designed specifically for parallel AI agents, aiming to replace human-centric workflows at the infrastructure layer. Origin is engineered to support the unique requirements of AI agents, facilitating more efficient and automated code management.

LlamaIndex advocates hybrid vector and grep retrieval for AI agents to achieve speed and precision

LlamaIndex is advocating for a hybrid retrieval architecture that combines semantic search (vector-based) with traditional grep-like keyword search for AI agents. This approach aims to achieve both speed and precision, addressing instances where vector-only methods might miss critical details in document retrieval.

LiteLLM Vulnerability: 6 CVEs and a Supply Chain Attack create systemic risk across 97M installations

Six critical CVEs, including a CVSS 10.0 RCE chain, have been discovered in the LiteLLM proxy, alongside a March supply chain compromise of its PyPI packages. These vulnerabilities pose systemic risks across 97 million monthly installations used by popular frameworks like CrewAI, DSPy, and Microsoft GraphRAG.

Cloudflare releases Flue 1.0 Beta, an open-source declarative agent framework for production AI agents

Cloudflare has released Flue 1.0 Beta, an open-source declarative agent framework that integrates with its Agents SDK. Flue provides durable execution, sandboxed code execution, and multi-cloud deployment capabilities for building production-grade AI agents.

Hugging Face open-sources ml-intern, an autonomous agent to automate ML research workflows

Hugging Face has open-sourced 'ml-intern,' an autonomous agent designed to automate machine learning research workflows. Since its beta launch, it has been used over 12,000 times to create hundreds of models by reading papers, implementing ideas, and iterating on model training.

Cloud & platform providers

AWS enters context layer market with self-learning knowledge graph for AI agents

Amazon Web Services (AWS) has entered the contested context layer market with a self-learning knowledge graph service designed to improve from agent usage without manual re-curation. This offering, including AWS Context and S3 Annotations, positions AWS against rivals like Snowflake Horizon and Microsoft Fabric IQ.

Amazon launches Web Search for Bedrock AgentCore, grounding AI agents in current web knowledge

Amazon has launched a new Web Search feature for Bedrock AgentCore, enabling AI agents to ground their responses in current and accurate web knowledge. This capability operates within a secured AWS environment, preventing data egress and enhancing the reliability of agent interactions.

AWS expands Bedrock AgentCore with native access to organizational, web, and paid content knowledge layers

AWS has expanded Bedrock AgentCore to include native access to three knowledge layers: organizational data, web search, and paid content. This update also adds continuous learning and monitoring capabilities, aiming to close gaps in agent performance and deployment for enterprises.

AWS Unveils Continuum: AI-driven agentic platform to detect and remediate code vulnerabilities at machine speed

AWS has unveiled Continuum, an AI-driven agentic platform that shifts from passive security dashboards to active vulnerability management. Continuum discovers, prioritizes, validates, and remediates code vulnerabilities at machine speed, incorporating graduated trust and human-in-the-loop oversight.

AWS unifies fragmented AI tooling into integrated agent orchestration platform

Amazon Web Services (AWS) is consolidating its fragmented AI tooling into a unified agent orchestration platform. This new platform integrates capabilities spanning data governance, code security, and business application execution, aiming to provide a more cohesive AI development and deployment experience.

Microsoft launches Copilot Cowork in Ireland as an agentic AI system for complex workflows

Microsoft has launched Copilot Cowork in Ireland, introducing it as an agentic AI system capable of executing complex multi-step workflows across Microsoft 365 applications. This move signifies a shift from mere AI tools to comprehensive task completion systems.

Cloudflare introduces Cloudflare One stack: agent-powered deployment for Zero Trust network configuration

Cloudflare has launched its Cloudflare One stack, an agent-powered skill set designed to automate Zero Trust network configuration, deployment, and migration. This new offering helps enterprises transition from legacy SASE vendors like Zscaler and Palo Alto Networks.

Cloudflare launches Cloudforce One threat events platform for real-time contextual threat intelligence

Cloudflare has launched Cloudforce One, a new platform designed to deliver real-time, contextual threat intelligence powered by its global network telemetry and AI. The platform aims to help security teams respond faster and more effectively to cyberattacks.

Cloudflare integrates Zero Trust identity controls with Ping Identity for edge-based AI agent enforcement

Cloudflare is positioning its global network as a control layer for AI workloads by integrating Zero Trust identity controls with Ping Identity. This partnership aims to enable edge-based enforcement for AI agents, enhancing security and governance for distributed AI applications.

Cloudflare launches new Design Partner tier and AI skills library, with Australian firm among first selected

Cloudflare has introduced a new Design Partner tier within its PowerUP program, providing selected global partners with technical training and financial support. This initiative also includes an AI skills library, the Cloudflare One Stack, to automate SASE deployment workflows, with an Australian firm among the first selected.

AI policy, regulation & governance

Anthropic, Google DeepMind, and OpenAI CEOs call for U.S.-led AI coalition at G7 summit

The CEOs of Anthropic, Google DeepMind, and OpenAI publicly called for U.S. leadership in forming an international AI standards and rules coalition during the G7 summit. They emphasized the risks posed by advanced cyber-capable models and the need for unified governance.

UK Government rolls out Extract planning automation tool nationwide, announces 2027 deployment for AI-powered APD

The UK Government is rolling out its Extract planning automation tool nationwide and has announced a 2027 national deployment timeline for the AI-powered Augmented Planning Decisions (APD) tool. Both tools are built on Google Cloud infrastructure and aim to streamline housing approval processes.

Emily Dalton Smith, head of Meta's AI restructuring, departs company amid workforce discontent

Emily Dalton Smith, who led Meta's AI restructuring initiative (Agent Transformation Accelerator), has departed the company. Her departure comes amid reported staff discontent over AI-driven workforce reductions and consolidation of internal tooling.

Japan has committed to 'continuously review' its AI laws in response to threats posed by high-performance models such as Claude Mythos. This signals an adaptive regulatory approach aimed at staying ahead of rapidly evolving AI capabilities and potential risks.

European Parliament backs AI Act simplification and ban on nudifier apps, addressing generative AI harms

The European Parliament has approved amendments to the EU AI Act, simplifying certain aspects and delaying compliance deadlines for some high-risk applications. Crucially, it added a ban on 'nudifier' apps and non-consensual intimate imagery tools, directly addressing fast-growing harms from generative AI.

Ireland establishes dedicated national AI regulator ahead of assuming EU Council presidency

Ireland has established a dedicated national Artificial Intelligence regulator, positioning itself at the forefront of implementing the bloc-wide EU AI Act. This move comes ahead of Ireland assuming the EU Council presidency, signaling a proactive approach to AI governance.

California state legislators partner on AI safety standards commission and auditor registry framework

California state legislators Senator Jerry McNerney and Assemblymember Rebecca Bauer-Kahan announced a partnership to establish a first-of-its-kind voluntary AI safety standards commission and an independent auditor registry framework. This initiative aims to promote responsible AI development within the state.

US export controls on Anthropic's AI models spark EU backlash over technological dependency

U.S. export controls on Anthropic's advanced AI models have provoked backlash within the EU, raising concerns about technological dependency. European officials are now decrying the American 'kill-switch' and renewing calls for greater digital sovereignty.

Australian Privacy Act amendments mandate AI disclosure for automated decision-making by December 2026

Amendments to the Australian Privacy Act will mandate AI disclosure for automated decision-making, with obligations taking effect on December 10, 2026. Non-compliance could result in penalties up to $50 million, compelling businesses to increase transparency around AI use.

US AI export shock exposes Australia's dependence on foreign AI, underscores need for sovereign infrastructure

Recent U.S. export controls on Anthropic's frontier AI models have starkly exposed Australia's heavy dependence on foreign AI suppliers. This situation underscores the urgent need for Australia to develop sovereign AI infrastructure and policies to secure its intelligence future.

Industry & market moves

Anthropic becomes first AI startup to join Frontier carbon removal coalition, pledging $915 million

Anthropic has joined Frontier, a carbon removal collective backed by major tech companies like Stripe and Google, becoming the first pure AI company to do so. The company has pledged $915 million, signaling a growing commitment from the AI sector to address environmental sustainability.

Anthropic opens official Seoul office, deepening ties with Korean AI ecosystem

Anthropic has established its first official office in Seoul, South Korea, to foster deeper partnerships with Korean enterprises, startups, and research institutions. This expansion aims to capitalize on the growing adoption of Claude across major Korean conglomerates, despite U.S. export controls on advanced AI models.

DXC Technology strikes global Claude AI alliance with Anthropic for mission-critical enterprise systems

DXC Technology has become Anthropic's Global Premier partner, entering into an alliance to deploy Claude into mission-critical systems across key industries such as banking, aviation, insurance, and government. The partnership includes training tens of thousands of certified engineers to facilitate these deployments.

Google Cloud announces multi-year partnerships with HSBC and Deloitte for UK enterprise AI adoption

Google Cloud has announced multi-year partnerships with HSBC and Deloitte to scale agentic AI adoption across UK enterprise and public sectors. These collaborations aim to drive the UK's transition from AI potential to agentic reality.

PM Modi and Mistral AI CEO Arthur Mensch discuss human-centric AI and India partnerships at G7

Indian Prime Minister Narendra Modi met with Mistral AI CEO Arthur Mensch at the G7 summit to discuss human-centric AI, open-weight models, and potential AI partnerships in India. The discussion signals India's interest in building sovereign AI capabilities and collaborating with European AI leaders.

Bell AI Fabric, Cohere, Hypertec, and BUZZ HPC announce landmark deal for Canadian sovereign AI

Four Canadian technology leaders—Bell AI Fabric, Cohere, Hypertec, and BUZZ HPC—have announced a landmark infrastructure deal to advance sovereign AI capabilities in Canada. The partnership combines Bell's data centers, Cohere's enterprise AI models, Hypertec's Canadian-built hardware, and BUZZ HPC's compute resources.

Robo.ai announces proposed acquisition of QC Capital, an AI-driven venture-building platform, for $60M

Robo.ai Inc. has announced a proposed acquisition of QC Capital Limited, an AI-driven technology holding and venture-building platform, for $60 million in equity consideration. The deal aims to strengthen Robo.ai's AI robotics ecosystem and venture incubation capabilities.

NeuralTrust raises $20M seed round, largest EU cybersecurity seed round, to secure enterprise AI agents

NeuralTrust has secured a $20 million seed round, marking the largest EU cybersecurity seed round to date. The funding positions the company to scale its AI agent security solutions as enterprises deploy thousands of autonomous agents into production environments.

CNTXT AI raises $60 Million Series A to scale sovereign AI infrastructure globally

CNTXT AI has secured a $60 million Series A funding round to scale its sovereign AI infrastructure globally. The company aims to enable enterprises and governments to develop AI solutions while maintaining full data sovereignty and control.

Behavox, an AI-native fintech controls platform, closes $175m equity raise from HPS Investment Partners

Behavox, an AI-native fintech controls platform, has closed a $175 million preferred equity round from HPS Investment Partners, a BlackRock affiliate. This marks its first major equity raise since SoftBank's $100 million investment in 2020, fueling its global growth ambitions.

Vaisala acquires AI weather company Atmo for $70M+ to lead meteorology transformation

Vaisala Corporation has acquired AI weather forecasting company Atmo, Inc. for a fixed sum of USD 70 million plus a potential USD 60 million earn-out. This acquisition aims to integrate frontier AI models with Vaisala's century-old sensor and observation networks, positioning it as a global leader in weather technology.

Accenture proposes acquiring Dragos, runZero, and NetRise for $4.175B to strengthen critical infrastructure defense

Accenture plans to expand into operational technology (OT) cybersecurity by proposing the acquisition of a majority stake in Dragos and full acquisitions of runZero and NetRise. This $4.175 billion combined deal aims to deliver end-to-end OT security for critical infrastructure amidst AI-driven cyber threats.

Qualcomm reportedly explores $8-10 billion acquisition of AI chip startup Tenstorrent

Qualcomm is reportedly exploring the acquisition of Tenstorrent, an AI chip startup known for its RISC-V technology, in a deal potentially valued between $8 and $10 billion. This move would significantly expand Qualcomm's position in AI computing.

Bull and Foxconn advance European AI manufacturing with NVIDIA Vera Rubin NVL72 platform production

Bull and Foxconn are advancing European AI manufacturing capacity by beginning production of NVIDIA Vera Rubin NVL72 systems in Europe. This partnership aims to serve AI factories and cloud providers with domestically manufactured AI infrastructure.

AI product & feature launches

Meta unveils new AI-driven commerce and discovery tools at Cannes 2026

Meta has launched a new AI-driven commerce suite at Cannes 2026, combining live shopping, virtual card checkout, and creator affiliate tools. This aims to streamline product discovery and purchase for its 3.5 billion users across Facebook, Instagram, and Messenger.

RAI to debut 'RAI Brain' cognitive architecture for humanoid robots with persistent identity and memory

Dutch AI company RAI Robotics Advanced Intelligence is set to debut 'RAI Brain,' a cognitive architecture designed to give humanoid robots persistent identity, autobiographical memory, and long-term intellectual development. This marks a significant departure from conventional episodic AI systems.

Foxconn debuts humanoid robots in Europe, showcasing vertically integrated physical-AI stack

Foxconn has made its first European debut of its vertically integrated physical-AI stack, showcasing humanoid robots with NVIDIA Isaac GR00T integration. This demonstration highlighted closed-loop control from AI compute (Vera Rubin) to robot training and physical deployment at scale.

Midjourney pivots to medical hardware, unveils full-body ultrasonic scanner with 50,000+ deployments planned

Midjourney, originally known for AI image generation, is pivoting into medical hardware with the unveiling of a full-body ultrasonic scanner. The company plans to deploy over 50,000 scanners by 2031, partnering with Butterfly Network and seeking FDA approval for a spa-based rollout strategy.

RealSense unveils AI-native D585 Pro depth camera with on-device AI for autonomous mobile robots

RealSense has unveiled the AI-native D585 Pro depth camera, the first with on-device AI processing, an edge inference engine, and a Gen 5 SoC. It delivers twice the depth quality and a sub-15cm minimum range, specifically designed for autonomous mobile robots and humanoids.

Pinterest launches 'Ask Pinterest' app and MCP Tools as AI reshapes product discovery

Pinterest has launched 'Ask Pinterest,' a new conversational AI app designed for product discovery, leveraging its proprietary Taste Graph. Concurrently, it has introduced MCP infrastructure for advertisers, indicating a deep integration of AI into its core business model.

Alchemy introduces AgentCard, a Visa-integrated payments and identity platform for AI agents

Alchemy has launched AgentCard, a new payments and identity platform that enables AI agents to autonomously make online purchases on behalf of consumers. Built on Visa Intelligent Commerce, it aims to provide secure and verifiable transactions for AI-driven services.

Google unveils Gemini-Powered Home Speaker to revive smart audio devices with natural multi-step commands

Google has launched its first new standalone smart speaker in six years, powered by Gemini AI. The device aims to revive the smart audio market by enabling natural multi-step voice commands and conversational interactions, moving beyond rigid command structures.

Research with immediate practical relevance

OpenAI releases LifeSciBench, a 750-task benchmark for grading AI models on real-life science research

OpenAI has introduced LifeSciBench, a new 750-task benchmark featuring expert-authored tasks and rubric-based grading to evaluate AI models on realistic life-science research workflows. This benchmark aims to measure AI performance on complex, multi-step scientific reasoning, addressing a gap in current evaluation methods.

OpenAI introduces Deployment Simulation research to spot AI behavioral drift before release

OpenAI has introduced Deployment Simulation, a novel pre-deployment testing method that replays real user conversations to detect AI behavioral drift. This research addresses the challenge of identifying problems that often emerge only in production environments, beyond traditional evaluation methods.

Google DeepMind unveils a framework to protect against its own rogue AI agents

Google DeepMind has publicly released a framework designed to safeguard against rogue AI agents within its own research operations. This initiative shares its approach with the broader AI research community, emphasizing security as AI capabilities grow.

Open standards (MCP, ARD, x402) emerge as foundational infrastructure for agent economy, shifting control to domain owners

Open standards such as Model Context Protocol (MCP), Agentic Resource Discovery (ARD), and x402 payment protocol are being widely adopted as the foundational infrastructure for the nascent agent economy. This trend is shifting control away from proprietary platforms towards domain owners and decentralized registries.

Two new medical AIs, MIRA and AMIE, match or exceed doctors in diagnosis and treatment decisions

New research published in Nature reveals two conversational AI systems, MIRA and AMIE (led by Google), demonstrating clinical reasoning and disease management capabilities that match or exceed physician performance. These AIs show high accuracy in diagnosis, treatment planning, and guideline adherence.

MIT builds DAAAM memory system, enabling robots to remember object locations through natural language queries

MIT researchers have developed DAAAM (Describe Anything, Anywhere, Anytime, at Any Moment), a memory system that combines computer vision and 3D mapping. This technology allows robots to maintain persistent long-term spatial memory and accurately respond to natural language queries about object locations.