The Informed Layer · Newsscan
Daily briefing

Anthropic Ode launch; regulation tightens

← All issues

Anthropic / Claude ecosystem

Anthropic, Blackstone, H&F launch Ode, a $1.5B enterprise AI services firm

Anthropic, Blackstone, and Hellman & Friedman introduced Ode, a new commercial AI services company that deploys anthropic frontier models inside enterprises. Funded with $1.5B, Ode aims to capture the implementation layer of AI, combining model access with professional services.

Claude Code can now invoke MCP connectors for richer artifacts

Anthropic added the ability for Claude Code Artifacts to securely access external tools and data sources via MCP connectors. This enables developers to build AI-powered applications that pull in live data from multiple services.

Frontier model providers

Google DeepMind and Isomorphic Labs launch bioresilience program

Google DeepMind and Isomorphic Labs announced a joint initiative to use AI, including Gemini, for biodesign and bioresilience, focusing on preventing, detecting, and responding to biological threats. Access will be provided to trusted partners with safeguards.

Google delays Gemini 3.5 Pro release to improve performance

Google has postponed the launch of its next-generation Gemini 3.5 Pro model, originally expected July 17, after internal benchmarks failed to meet targets. The company is rebuilding parts of the model to strengthen coding and reasoning capabilities.

Google's Gemma 4 gets tool calling and vision improvements

The latest Gemma 4 update introduces Flash Attention 4 support, enhanced tool calling reliability, and improved vision capabilities, including larger token budgets for OCR. These fixes boost agentic performance on edge devices.

SpaceXAI open-sources Grok Build: Rust agent harness and TUI

Elon Musk's SpaceXAI released the core Rust-based agent harness, terminal user interface, and tool layer behind its Grok Build CLI, allowing developers to inspect and customize the system.

Moonshot AI launches Kimi K3, largest open-source model at 2.8T parameters

China's Moonshot AI officially released Kimi K3, a 2.8-trillion-parameter open-source model, claiming it rivals top US systems. The release follows a leak earlier in the week.

AI developer tooling & infrastructure

Atlassian launches AI-native development system in Jira

Atlassian announced a new system embedding AI agents directly into Jira to plan, orchestrate, and govern AI-driven software development across the entire SDLC. The system includes a Jira Coding Agent, Planner, and integrations with Slack and Loom.

Internet Identity preps MCP server to let AI agents act across apps securely

A new MCP server under development aims to allow AI agents to perform actions across applications without handling user credentials, leveraging decentralized identity to standardize secure multi-provider integration.

Cloud & platform providers

Google Cloud adds grounding with Parallel Web Search on Gemini Enterprise

Google Cloud introduced a new Grounding with Parallel Web Search feature for Gemini Enterprise agents, allowing them to retrieve real-time, citeable information from the web during agentic tasks.

Amazon EC2 G7e instances now available in more regions

AWS expanded availability of its EC2 G7e instances, powered by RTX PRO 6000 Blackwell Server Edition GPUs, to additional regions including Europe. The instances offer up to 2.3x inference performance over previous generation.

AI policy, regulation & governance

Japan revises AI policy guidelines to strengthen cybersecurity

Japan's Cabinet updated national AI policy guidelines to address cybersecurity risks from rapid AI advances, calling for international cooperation and domestic development of secure AI systems.

Hawaii enacts laws to protect children from AI deepfakes and chatbots

Hawaii Governor signed two bills into law: SB 3001, the Artificial Intelligence Disclosure and Safety Act, requiring age-appropriate disclosures for AI chatbots, and HB 2137, establishing civil protections against AI-generated deepfakes targeting minors.

Germany classifies AI search and chatbots as content providers under media law

Germany's media authorities (ZAK) issued a first-of-its-kind ruling placing AI-powered search and conversational products like Google AI Overviews and Perplexity under the State Media Treaty, requiring compliance with regulatory standards for content providers.

Australia to introduce AI and data centre laws in regulatory pivot

The Australian government announced plans to legislate national AI standards, including a mandatory data centre framework to regulate energy use and compensate creative industries, moving beyond the recently outlined national framework.

Industry & market moves

University of Toronto and Cohere partner to scale AI adoption

The University of Toronto and Cohere announced a strategic partnership to advance scalable, responsible AI adoption across the university, focusing on safety and data sovereignty. The collaboration includes joint research and deployment of Cohere's models.

CI&T and Mistral partner to build private AI stacks for enterprises

Global IT services firm CI&T named Mistral AI as its preferred AI partner, striking a multi-year deal to deploy open-weight models and end-to-end private AI stacks across enterprises, with an emphasis on Latin America.

Chinese regulatory filing implies DeepSeek valuation around $52 billion

A new corporate filing in China reveals that a fund acquired a 0.8265% stake in DeepSeek for 2.9 billion yuan, implying a valuation of approximately 350.88 billion yuan ($51.82 billion). This provides a concrete data point as the company seeks an IPO.

3M and Microsoft strike partnership for AI data center infrastructure

3M and Microsoft announced a strategic collaboration where Azure will deploy 3M's Expanded Beam Optical technology in data centers to improve AI infrastructure efficiency, while 3M uses Microsoft's AI platforms for enterprise transformation.

Intel and Google Cloud expand collaboration for AI-powered enterprise transformation

Intel and Google Cloud announced a broadened multi-year partnership to accelerate Intel's internal AI transformation using Gemini Enterprise and Google Cloud, including AI-driven silicon design and scalable workflows.

Hyundai Motor Group takes full control of Boston Dynamics to accelerate physical AI

Hyundai Motor Group is acquiring the remaining stake in Boston Dynamics to fully own the robotics pioneer, aiming to speed commercialization of physical AI and advanced robot platforms.

Microagi raises $55M seed to teach factory robots using telemetry data

Munich-based Microagi secured a $55 million seed round led by Hummingbird to build Atlas, a data platform that trains factory robots by capturing human telemetry and adapting existing models.

Whatnot acquires AI startup Shaped to boost live shopping recommendations

Live shopping platform Whatnot acquired Shaped, an AI startup specializing in real-time recommendation engines, to enhance product discovery during livestreams.

Instacart acquires Arpalus for real-time shelf intelligence

Instacart acquired Arpalus, a computer vision startup, to integrate real-time shelf monitoring into its fulfillment and store operations, enhancing inventory accuracy for grocery retail.

BandLab acquires Israeli AI music startup Aiode

BandLab Technologies acquired Aiode, an AI-powered music creation tool, in a deal estimated at tens of millions of dollars, reinforcing its creative AI portfolio.

Legal AI platform Harvey acquired Benchmark, a firm specializing in AI for asset management, signaling a move to broaden its enterprise footprint into financial services.

AI product & feature launches

Microsoft Scout enhances AI model and chat navigation flexibility

Microsoft upgraded its Scout Autopilot feature with user-selectable context windows, per-message model switching, and expanded data-source access, aiming to improve developer productivity when using AI copilots.

Sakana AI integrates Nvidia Nemotron into its collective intelligence orchestrator

Sakana AI added Nvidia's Nemotron model to its Fugu orchestration system, combining multiple open models to rival the performance of single frontier models through collaborative intelligence.

Xiaomi releases Robotics-1 base model for embodied AI

Xiaomi announced Xiaomi-Robotics-1, a base model for robots that leverages large-scale pre-training and post-training data curation, designed for out-of-the-box usability in robotics applications.

PrismML compresses 27B model to fit on a phone in 3.9GB

PrismML announced its Bonsai 27B model can now run on smartphones after aggressive compression to 3.9GB, preserving roughly 95% of full-precision accuracy on benchmarks.

Nvidia launches Cosmos 3 Edge AI model for real-time robotics

Nvidia introduced Cosmos 3 Edge, a version of its omnimodel optimized for edge deployment in robotics and vision AI. The model is part of a broader push with Japanese industry to build national physical AI infrastructure.

Japan, industrial leaders, and NVIDIA launch world's first national AI infrastructure

In a joint initiative with the Japanese government, companies including NVIDIA will build a large-scale AI factory based on Vera Rubin chips, managed by Noetra Corp, to serve as a national foundation for physical AI development.

Research with immediate practical relevance

Security researchers discovered that xAI's Grok Build command-line tool was secretly uploading entire codebases to Google Cloud Storage without user permission, triggering a server-side disablement and data retention concerns.

WANDA generates infinite training data for robot manipulation from a single demo

Researchers introduced WANDA, a synthetic data engine that creates thousands of diverse training trajectories from one human demonstration, dramatically reducing the data collection burden for teaching robots mobile manipulation skills.

Single prompt can trigger ChatGPT to execute full cyber-attack chain, researchers claim

A security study claims that by crafting a single prompt, threat actors can cause ChatGPT-5.5 to autonomously carry out a complete cyber-attack sequence, including reconnaissance, exploitation, and data exfiltration.