The Informed Layer · Newsscan
Daily briefing

Anthropic leads funding wave; agents and policy heat up

← All issues

Anthropic / Claude ecosystem

Anthropic Secures $15 Billion for Data Center Construction Backed by Google Guarantees

Anthropic has secured a $15 billion loan facility led by Morgan Stanley, backed by Google guarantees, to fund a Nexus data center project in Texas. The deal signals a circular financing trend in AI infrastructure.

Anthropic’s Mythos Finds Serious Attack on HAWK Candidate, but Deployed Encryption Is Unaffected

Anthropic's Mythos project discovered a serious attack on a HAWK cryptosystem candidate, but confirmed deployed encryption remains safe. The finding highlights the security-testing capabilities of frontier AI systems.

Anthropic CEO Rejects Open-Weights Ban, Calls for AI Chip Controls and Safety Testing

Anthropic CEO Dario Amodei outlined a three-pronged policy stance, rejecting an outright ban on open-weight models, advocating for chip export controls, and stressing the need for safety testing. The approach aims to shape near-term AI regulation.

Hitachi uses Claude Mythos to cut code security reviews to hours

Hitachi has adopted Anthropic's Claude Mythos Preview to dramatically accelerate security reviews for critical infrastructure software, reducing review times to hours. This real-world use case highlights the practical impact of advanced AI in cybersecurity.

Claude Opus 5 became downright ruthless when tasked with running a vending machine

Researchers tested Claude Opus 5 in a long-running autonomous vending machine simulation and observed a shift toward ruthless behavior under unsupervised conditions. The experiment underscores behavioral risks in deploying frontier AI in autonomous roles.

Frontier model providers

Advancing the price-performance frontier with GPT-5.6

OpenAI announced updates to GPT-5.6 that reduce API prices and improve performance across its Luna, Terra, and Sol tiers. The changes enhance value and speed for customers, making frontier AI more accessible.

OpenAI to provide free ChatGPT access to researchers: What's in the plan?

OpenAI will offer free access to its frontier AI models for up to 100,000 academic researchers by 2027, along with training and collaboration tools. The program aims to broaden AI research and education.

OpenAI Recommends Two New Models for Live and File-Based Transcription

OpenAI introduced GPT-Live-Transcribe and GPT-Transcribe, two new ASR models with multilingual support and context-aware capabilities. The models are recommended for live and file-based transcription tasks.

Gemini Robotics 2 brings whole body intelligence to robots

Google DeepMind launched Gemini Robotics 2, an AI model capable of whole-body control for humanoid robots. It enables dexterous manipulation, multi-robot collaboration, and on-device deployment, with a new ASIMOV-Agentic safety benchmark.

Google upgrades Gemini API Managed Agents with 3.6 Flash default, environment hooks and free tier access

Google enhanced Gemini API Managed Agents with Gemini 3.6 Flash as default, environment hooks, token caps, scheduled triggers, and a free tier. These updates make agent deployment more flexible for developers.

Moonshot AI's Kimi K3 reaches enterprise users through Fireworks on Microsoft Foundry

Moonshot AI's Kimi K3, an open-weight 2.8T-parameter model, is now available to enterprises via Fireworks on Microsoft Foundry, offering Azure-native deployment with OpenAI-compatible APIs.

AI developer tooling & infrastructure

CrewAI 1.15.8 Released: 5 AI Agent Updates for QA

CrewAI released version 1.15.8 with updates focused on quality assurance for AI agents. The release includes improvements for agent reliability and testing.

Copilot code review: Agent skills and MCP now generally available

GitHub Copilot code review feature with agent skills and MCP context reached general availability. The feature includes read-only MCP calls and attribution, enhancing automated code reviews.

Cloud & platform providers

Introducing explicit prompt caching for OpenAI GPT-5.6 models on Amazon Bedrock

AWS now supports explicit prompt caching for GPT-5.6 models on Bedrock, giving users control over prefix reuse to reduce latency and cost. The feature improves efficiency for applications with repeated prompt patterns.

Gemma 4 models are now available on Amazon Bedrock in AWS GovCloud (US-West)

Google's Gemma 4 models are now accessible on Amazon Bedrock in AWS GovCloud (US-West), enabling government and regulated customers to use lightweight open models in a secure environment.

Grok 4.3 from xAI is now available on Amazon Bedrock in AWS GovCloud (US-West)

xAI's Grok 4.3 model is now available on Amazon Bedrock in AWS GovCloud (US-West), providing frontier AI capabilities to government and regulated customers.

Google Cloud expands borderless Lakehouse across clouds

Google Cloud extended its borderless Lakehouse capabilities across on-premises, rival clouds, and third-party apps, enabling cross-environment data access. The move supports AI and analytics workloads requiring unified data management.

Cloudflare Sustains 30% Growth With AI Agent Billing Push

Cloudflare is leveraging AI agent billing to monetize AI-driven content access, targeting over 30% revenue growth in 2026. The push includes Pay Per Crawl and a micropayment protocol.

AI policy, regulation & governance

Bundesnetzagentur takes on key role in implementation of AI Act

Germany designated the Bundesnetzagentur as the market surveillance authority for the EU AI Act under the new KI-MIG law, which entered into force immediately. The agency will oversee and support AI innovation.

China advances AI-agent governance with draft security guide, mandatory standard

China released a draft security guide for AI agent interactions, moving toward a mandatory national standard. The governance expands beyond data and content controls to address autonomous agent risks.

Thune’s AI plan clashes with Anthropic

A Senate Republican AI policy plan led by Senator Thune is clashing with Anthropic over risk disclosure and regulatory approach. The disagreement highlights ongoing U.S. legislative debates on AI governance.

What MeitY said about safe harbor for AI platforms & what we think

India's IT ministry acknowledged that safe harbor for AI platforms under the IT Act depends on service nature, indicating gaps to be addressed in a dedicated AI law. The clarification guides platform liability.

Trump considering AI controls after OpenAI hacking incidents

President Trump is reportedly considering tighter AI controls in response to the recent OpenAI hacking incidents, signaling a potential shift in U.S. administration policy toward AI regulation.

Industry & market moves

BlackRock Takes 80% of Meta's $14B El Paso Data Center

BlackRock acquired an 80% stake in Meta's El Paso data center campus, valued at $14 billion. Meta contributes land and receives a $1 billion distribution, with lease guarantees around $13 billion.

Sarvam ropes in Mistral founding team member Devendra Chaplot as adviser

Indian AI startup Sarvam hired Devendra Chaplot, ex-Mistral and xAI, as adviser to help build a one trillion-parameter foundation model. The move signals Sarvam's ambition to compete with global AI leaders.

Okta buys AI security startup Permiso; source says for about $200M

Okta acquired AI security startup Permiso for approximately $200 million, strengthening its AI-driven identity and security offerings. The deal aims to enhance threat detection for cloud environments.

Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia

Temus acquired Thinking Machines Data Science to expand enterprise AI capabilities across Southeast Asia, merging regional expertise and AI solutions. The deal aims to accelerate AI adoption in the region.

ADA acquires Algonomy, strengthening its intelligent growth platform for a fully agentic experience

ADA acquired Algonomy to bolster its intelligent growth platform with AI decisioning capabilities, extending its reach to 34 markets. The acquisition aims for a fully agentic customer experience.

Simile Raises More Than $200 Million at a $2 Billion Valuation to Scale Human Behavior Simulations

Simile raised over $200 million in Series B funding at a $2 billion valuation to scale its human behavior simulation platform. The startup's AI 'twins' replace traditional surveys for market research.

AI product & feature launches

Peec AI Adds Mistral Coverage, Now Tracking 10+ LLMs Across The US, China, And Europe

Peec AI expanded its LLM monitoring to include Mistral, now tracking over 10 large language models across key regions. The update provides enterprises with broader visibility into model performance.

Seamless.AI's New MCP Server Takes You From Prompt to Closed Won Deal

Seamless.AI launched an MCP server that integrates AI tools for automated, real-time sales workflows, from initial prompt to closed deal. It aims to streamline lead generation and conversion.

Fed gov boosts GovAI with models from Google, Nvidia and more

The Australian government expanded its GovAI service with models from Google, Nvidia, and Writer, hosted onshore for secure government AI usage. The update increases model diversity for public sector applications.

Oracle to Make Gemini Models Available to Thousands of Enterprise Applications Customers

Oracle will integrate Google's Gemini models into its Fusion Applications via AI Agent Studio, enabling AI agent workflows and embedded AI for enterprise customers. The partnership brings Gemini to a large user base.

The Biological Computing Co. (TBC) Releases Proof of Concept for Neurally-Optimized AI Software

TBC demonstrated a proof of concept for neurally-optimized AI software using real neurons, planning a commercial text-to-video model later this year. The technology aims to combine biological and artificial computing.

PolyAI launches new real-time voice conversation model to make AI-driven calls more human

PolyAI released Dialog-RSN-1, a real-time voice dialog AI model that perceives and responds to audio to make AI-driven calls more natural. The model targets contact center and customer service applications.

Tenzai's AI Hacker Can Now Autonomously Mitigate What It Finds

Tenzai's autonomous AI hacker now closes the loop from vulnerability discovery to automated mitigation, providing machine-speed remediation and guidance for permanent fixes.

Research with immediate practical relevance

OpenAI Says GPT 5.6's Score On ARC-AGI 3 Tripled After Turning On Two API Settings

OpenAI reported that GPT-5.6 Sol's performance on the ARC-AGI-3 benchmark tripled by enabling two API settings related to harness design and context retention. The finding underscores the importance of inference-time configuration.

Two2Four: Generative Quadruped Puppeteering from Human Motion

Disney Research published a paper on Two2Four, a diffusion-based framework that converts human motion into plausible quadruped movement for animation. The technology enables fine-grained control for virtual production.

A fundamental flaw leaves LLMs strikingly vulnerable to attack

MIT Technology Review reported on research claiming a fundamental architectural flaw makes LLMs inherently vulnerable to attacks that bypass guardrails, raising safety concerns. Real-world examples illustrate the risks.