The Informed Layer · Newsscan
Daily briefing

Anthropic signs $35B deal; EU regulates ChatGPT

← All issues

Anthropic / Claude ecosystem

Anthropic Releases Claude Code v2.1.252 CLI Update

Anthropic shipped version 2.1.252 of Claude Code, resolving issues with Remote Control session stalling and background task output limitations while refining persistent project configurations. The patch improves stability for long-running automated developer sessions.

Frontier model providers

Google Releases TimesFM-3 Foundation Model for Multivariate Time-Series Forecasting

Google launched TimesFM-3, a 330-million parameter zero-shot foundation model pre-trained on more than 1 trillion data points for multivariate time-series forecasting. Unlike predecessor models, TimesFM-3 natively incorporates exogenous variables and correlated series without task-specific fine-tuning.

Meta Superintelligence Lab Debuts Muse Voice Transcribe

Meta's Superintelligence Lab announced Muse Voice Transcribe, expanding its Muse family of generative models into real-time speech-to-text processing. The tool offers low-latency transcription optimized for integrated multimodal agent workflows.

DeepSeek Open-Sources V4-Flash-Vision-Exp Weights for Local Deployment

DeepSeek released open weights for its 305-billion parameter V4-Flash-Vision-Exp multimodal foundation model under a permissive commercial license. The release enables enterprise developers to host and run vision-language agent pipelines on local infrastructure without external API dependencies.

AI developer tooling & infrastructure

Google Launches Gemini Multi-Agent Teamwork Framework in Antigravity

Google introduced Gemini Multi-Agent Teams within its Antigravity environment, powered by Gemini 3.7 Flash. The multi-agent orchestration architecture was demonstrated solving seven open mathematical conjectures and synthesizing complex hardware simulations.

Taiwan Central News Agency Launches MCP Data Connector for AI Agents

Taiwan's state-backed Central News Agency (CNA) introduced a subscription-based Model Context Protocol (MCP) server. The service delivers authenticated, structured news feeds directly into AI agents to mitigate hallucination risks in regional information queries.

Tealium Introduces Configuration MCP for Autonomous Customer Data Platforms

Tealium expanded its enterprise agent integration with the launch of the Tealium Configuration MCP. The connector allows authenticated AI agents operating across Claude, ChatGPT, or custom platforms to inspect and modify CDP schemas under strict administrative guardrails.

Legal technology provider FRAIM launched an MCP server for its LAWGUE platform. The tool enables legal teams to invoke proprietary contract repositories, clause analysis, and drafting verification directly within frontier AI assistants like Claude and ChatGPT.

Wantedly Launches Hire MCP Server for Automated HR Agent Queries

Recruitment platform Wantedly rolled out Hire MCP Server, enabling external conversational agents to query talent pipelines, candidate profiles, and interview records using structured permissions.

EvoMap Open-Sources AutoResearch for Autonomous Scientific Experimentation

EvoMap open-sourced AutoResearch, an agent framework designed to autonomously formulate research hypotheses, write code to execute experiments, evaluate empirical outcomes, and refine theories without human intervention.

Cloud & platform providers

AWS Introduces Certified AI Business Strategist Credential

Amazon Web Services launched the AWS Certified AI Business Strategist program, establishing an industry credential specifically tailored to non-technical executives and enterprise managers overseeing generative AI implementations.

AWS Launches Graviton5-Powered R9g Instances for High-Memory AI Workloads

AWS rolled out memory-optimized Amazon EC2 R9g and R9gd instances powered by Graviton5 custom silicon. The instances deliver up to 25% higher compute performance and 35% faster ML inference compared to prior Graviton generations.

AI policy, regulation & governance

Debian Project Rejects AI LLM Ban in Favour of Voluntary Disclosure

The Debian Project membership concluded a general resolution vote, rejecting proposals to ban AI-generated contributions to the open-source distribution. The project chose instead to encourage voluntary disclosure while relying on traditional peer code review maintainership standards.

US Urges G20 to Adopt 'Carolina Principles' Limiting Centralised AI Governance

The White House Office of Science and Technology Policy formally submitted the 'Carolina Principles' to the G20, advocating for sector-by-sector regulatory frameworks rather than establishing centralized multilateral AI oversight institutions.

EU Designates ChatGPT as Very Large Online Search Engine Under DSA

The European Commission formally designated OpenAI's ChatGPT as a Very Large Online Search Engine (VLOSE) under the Digital Services Act. The decision subjects OpenAI to strict transparency, researcher data access, and annual systemic risk audit mandates backed by fines up to 6% of global revenue.

US Lawmakers Propose Conditional AI Token Excise Tax Bill

Members of the US Congress introduced draft legislation proposing a conditional excise tax on commercial AI compute tokens and revenues. The bill proposes that tax rates automatically trigger or increase if domestic unemployment metrics rise due to technological displacement.

Australian Privacy Act Exposure Draft Imposes 72-Hour Breach Reporting Deadline

Details from the Australian Government's Privacy Amendment exposure draft reveal a mandatory 72-hour notification deadline for data breaches involving AI systems and digital platforms, replacing the legacy 'as soon as practicable' standard to align with international privacy regimes.

Australian Treasury Warns Lagging Enterprise AI Adoption Threatens Economic Growth

An Australian Treasury analysis warned that enterprise AI adoption remains below 10% across domestic businesses, creating risks that Australia will fall behind peer economies in productivity and capital attraction unless private sector implementation accelerates.

Pentagon Deploys Custom ChatGPT Mil and Grok on GenAI.mil to 1.7M Users

The US Department of Defense deployed specialized government editions of OpenAI's ChatGPT and xAI's Grok to 1.7 million defense personnel via its secure GenAI.mil infrastructure. Anthropic remains excluded from the initial deployment following previous supply-chain risk disputes.

Industry & market moves

Chinese AI Startup Manus Resumes Independence as Meta Deal Collapses

Chinese agent startup Manus has resumed independent operations after Chinese regulatory authorities blocked Meta's planned multibillion-dollar acquisition on national security grounds. The collapsed transaction reflects increasing geopolitical resistance to cross-border AI M&A.

Saudi AI Champion HUMAIN Unveils $15B in Infrastructure and Model Alliances

At the LEAP 5 conference, Saudi Arabia's HUMAIN announced over $15 billion in strategic deals, including a $5 billion AI Zone infrastructure deployment with AWS and autonomous trucking initiatives with Applied Intuition. The entity also launched the HUMAIN M3 Arabic foundation model suite.

NVIDIA made a $3.5 billion convertible investment in MediaTek to co-develop custom cloud AI accelerators embedded with NVIDIA's NVLink Fusion interconnect. The agreement enables hyperscalers to deploy semi-custom silicon tightly integrated with NVIDIA rack architectures.

AI Agent Security Startup AIR Raises $50M

Cybersecurity startup AIR emerged from stealth with $50 million in funding to build inline inspection firewalls for autonomous AI agents. The platform evaluates tool calls, skills, and third-party data inputs to prevent prompt injections and unauthorized agent executions.

Kuaishou's Kling AI Secures RMB1.4B From China National AI Fund

Kuaishou's generative video unit Kling AI closed a RMB1.4 billion (~$195M) funding round backed by China's state AI fund. The capital will fund large-scale video model pretraining and international expansion.

Anthropic Secures $35B Dedicated Compute Capacity Deal With Lambda

Anthropic signed a $35 billion long-term cloud infrastructure contract with NVIDIA-backed Lambda to secure massive dedicated GPU cluster capacity across US data centers through the end of the decade.

Visko Launches Orbis Interactive Video Foundation Model With $10M Funding

AI startup Visko emerged with $10 million in pre-seed backing and released Orbis, an interactive 4K video generation model capable of continuous, physics-consistent real-time scene rendering over extended horizons.

AI product & feature launches

NVIDIA Rolls Out DLSS 5 Neural Graphics Technology

NVIDIA officially debuted Deep Learning Super Sampling 5 (DLSS 5), integrating advanced neural rendering to reconstruct frames and lighting with claimed multi-fold performance improvements in commercial titles such as NBA 2K27.

Phonely Releases Alma Ultra-Low-Latency Voice Model

Phonely unveiled Alma, a conversational voice AI model pre-trained on 10 million real phone conversations. The architecture achieves 185ms first-token latency, substantially reducing conversational latency in automated voice telephony.

Flower Labs Releases Endeavor 1.0 On-Premises Frontier Model

Cambridge University spinout Flower Labs launched Endeavor 1.0, an open, federated foundation model engineered for on-premises deployment in regulated European enterprises seeking alternatives to US closed-cloud APIs.

John Deere Launches JD AI Assistant for Farm Data Operations

John Deere rolled out the JD AI Assistant, an integrated conversational agent embedded in its Operations Center platform that allows farmers to analyze historical yield, field telemetry, and machinery metrics with strong data privacy guarantees.

Reducto Launches r-1 Foundation Model for Complex Document Parsing

Reducto unveiled r-1, a specialized vision-language foundation model tailored for complex document extraction. The model claims a 20% reduction in parsing errors across complex tables and forms at an inference cost of one cent per page.

Optimizely Unveils Domain-Specific Marketing Models and Mark-Bench

Optimizely released a family of post-trained domain models optimized for enterprise marketing workflows, claiming 10x cost reductions compared to general-purpose frontier LLMs, alongside Mark-Bench, an open-source evaluation benchmark.

Apodex Launches Apodex 1.1 Multi-Agent Execution Platform

Apodex released Apodex 1.1 and the FrontierAgent framework, combining multi-agent reasoning, continuous execution workspaces, and cryptographic step verification for complex professional and scientific workflows.

Research with immediate practical relevance

Japanese Researchers Develop AF3-ReD to Uncover Alternate Protein Conformations

Researchers at Japan's Institute for Molecular Science developed AF3-ReD, applying repulsive bias within AlphaFold3's diffusion model to sample transient and dynamic protein states. Default AlphaFold models traditionally predict static equilibrium structures, limiting drug discovery applications.

Tsinghua AIR Unveils Zeva Embodied AI With In-Context Causal Learning

Tsinghua University's Institute for AI Industry Research (AIR) introduced Zeva, an embodied robotics intelligence framework that adapts and learns from physical interaction in real time using in-context causal learning without updating model weights.