The Informed Layer · Newsscan
Daily briefing

Agents and MCP surge; US urges light AI rules

← All issues

Anthropic / Claude ecosystem

Claude Code CLI 2.1.252 ships with bug fixes

Anthropic released a new patch for Claude Code CLI with multiple bug fixes aimed at improving stability and persistence during long agent sessions.

Anthropic finalizes $35 billion cloud deal with Lambda

Anthropic finalized a $35 billion cloud-computing agreement with Nvidia-backed provider Lambda, expanding AI training and inference capacity outside the largest hyperscalers.

Frontier model providers

Meta's Muse Image agentic image model available on fal

Meta Superintelligence Labs' Muse Image is now available on fal. The model plans, calls tools, and self-corrects within a single workflow to generate and edit images with higher accuracy.

Pentagon launches GenAI.mil with ChatGPT Mil and Grok

The US Department of Defense deployed custom secure versions of ChatGPT and Grok on GenAI.mil, making frontier generative AI tools available to up to 3 million personnel.

AI developer tooling & infrastructure

Japanese legal tech provider FRAIM launched an MCP server for LAWGUE, allowing customers to integrate contract data with their own generative AI tools for end-to-end contract review and editing.

EvoMap open-sources AutoResearch for AI agent experimentation

EvoMap released AutoResearch, an open-source system that lets AI agents propose, test, and refine their own research ideas through executable experiments and evidence-based decisions.

Reducto unveils r-1 frontier document parsing model

Reducto released r-1, a frontier parsing model that reduces errors up to 20% and makes complex documents AI-ready at 1 cent per page.

Taiwan's CNA launches MCP news data service

Central News Agency launched a subscription-based Model Context Protocol service that gives AI applications access to trusted Taiwan news, translations, and public data with clear sourcing.

Cloud & platform providers

AWS launches Certified AI Business Strategist certification

AWS introduced a portable business-judgment certification designed for people scaling AI initiatives across organizations.

AI policy, regulation & governance

US urges G20 to adopt Carolina Principles limiting AI rules

The US government proposed Carolina Principles at the G20, arguing that only genuinely novel regulatory gaps should be filled and existing laws used where possible.

Australian Treasury warns country risks missing global AI boom

Australia's Treasury issued a warning that the country may miss out on the global AI boom, signaling policy and investment concerns.

Debian developers reject LLM ban and make AI disclosure voluntary

The Debian project decided not to ban LLM use by contributors and will encourage but not require disclosure of AI assistance in code contributions.

ChatGPT becomes first AI chatbot under EU DSA strict rules

EU regulators designated ChatGPT as a Very Large Online Search Engine, subjecting it to the Digital Services Act's strictest transparency and risk-management obligations.

US bill proposes AI token tax tied to unemployment

Lawmakers introduced a dynamic excise tax on AI companies that would rise as unemployment increases, funding jobs in infrastructure and care services.

Industry & market moves

AIR raises $50M to vet AI agent skills and add-ons

AI security startup AIR raised a $50M round to build continuous vetting and a marketplace of trusted add-ons for AI agents, addressing supply-chain risk.

HiBob secures investment to fuel AI-powered enterprise

HR platform HiBob announced a new investment to expand its role as an organizational intelligence layer, grounding AI in trusted workforce context.

OpenAI rejects Apple trade-secret claims in court filing

OpenAI disputed Apple's allegations of trade-secret theft, arguing the dispute stems from Apple's own internal security and offboarding practices. The legal fight could shape talent movement between major AI and consumer hardware firms.

Arabic.AI secures strategic investment from HUMAIN

Arabic.AI and Tarjama received a strategic investment from HUMAIN, enabling enterprise-scale Arabic and multilingual AI deployment in Saudi Arabia and beyond.

NVIDIA invests $3.5 billion in MediaTek for custom AI chips

NVIDIA announced a $3.5 billion investment in MediaTek to expand custom AI accelerator development, with NVLink Fusion as the foundation for workload-specific XPUs.

Kling AI brings in RMB1.4 billion from China AI fund

Kuaishou's Kling AI unit secured RMB1.4 billion from a state-backed China AI fund and a strategic investment from CP Robot, reshaping ownership after its spinout.

AI product & feature launches

Tealium expands agentic capabilities with Configuration MCP

Tealium introduced a new Model Context Protocol configuration that lets AI agents and developers create, modify, and act on customer data with governance and consent controls.

Wantedly launches Hire MCP Server for AI-assisted recruiting

Wantedly released an MCP server for its Wantedly Hire recruitment management system, enabling secure connections between recruiting data and external AI agents.

NVIDIA DLSS 5 debuts in NBA 2K27 with pixel-accurate graphics

NVIDIA's DLSS 5 uses 3D-guided neural rendering to deliver up to sevenfold performance gains and pixel-accurate visuals in NBA 2K27, with developer controls over enhancement strength.

Visko closes $10M pre-seed and launches Orbis live model

AI startup Visko closed a $10 million pre-seed round and launched Orbis, its first live AI model, signaling a move toward running-in-world AI paradigms.

Phonely launches Alma voice AI trained on 10 million calls

Phonely released Alma, a voice AI model trained on over 10 million real phone conversations to improve natural conversational flow for voice agents.

Cambridge spinout launches Endeavor 1.0 model competitive with frontier labs

Flower Labs launched Endeavor 1.0, a frontier-class AI model that can run on customer premises and is competitive with offerings from OpenAI and Anthropic.

John Deere launches AI assistant for farmers

John Deere introduced JD, an AI assistant embedded in its Operations Center that answers operational questions using farmers' own data, with near-term rollout including in-cab displays.

Optimizely launches purpose-built marketing AI models

Optimizely introduced post-trained, purpose-built AI models for marketing that deliver frontier-level quality at a fraction of the cost, and released Mark-Bench, an open marketing AI benchmark.

Apodex 1.1 adds verifiable execution to AI research

Apodex released version 1.1 with a verifiable, end-to-end workspace that traces conclusions to sources and supports long-running, multi-step professional tasks with integrity checks.

fal launches H3 Max video generation model

fal released H3 Max, a post-trained video model with frontier quality and faster-than-real-time generation, outperforming comparable models in independent benchmarks.

Research with immediate practical relevance

Google unveils TimesFM-3 multivariate forecasting model

Google released TimesFM-3, a zero-shot multivariate forecasting foundation model that predicts across multiple data series such as sales and weather without additional training. It can forecast the target window all at once.

New AlphaFold3 sampling method predicts protein shape changes

Researchers introduced a repulsive bias scheme called AF3-ReD that forces AlphaFold3 to sample alternative protein conformational states beyond a single predicted structure. The approach expands AlphaFold3's use for studying dynamic protein behavior.

Google Antigravity with Gemini 3.7 Flash solves seven open problems

Google reported seven new theoretical and computational results from its Antigravity multi-agent system powered by Gemini 3.7 Flash, including a Lean-verified proof of a case of Knuth's Cycles Conjecture.

Tsinghua AIR unveils Zeva embodied AI system

Researchers introduced Zeva, an embodied AI system that uses in-context causal learning to improve from experience without weight updates, offering a new scaling path for robots.