The Informed Layer · Newsscan
Daily briefing

Frontier models ship; agents drive infra & regulation

← All issues

Anthropic / Claude ecosystem

UST, Anthropic expand Enterprise AI collaboration

UST and Anthropic are expanding their strategic alliance to embed Claude across various enterprise platforms, including engineering, healthcare, banking, and telecom. The collaboration also includes training 20,000 UST employees globally to facilitate broader adoption of trusted AI solutions.

CISA Deploys Anthropic’s Mythos AI to Hunt Vulnerabilities in U.S. Government Code

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has deployed Anthropic's Mythos AI to scan federal code repositories for vulnerabilities. This operational deployment marks a significant improvement in U.S. government-Anthropic relations following a previous blacklisting by the Pentagon.

Frontier model providers

How Google DeepMind’s new AI models can improve vision and reasoning in robots

Google DeepMind has introduced a two-model system, Gemini Robotics-ER 1.5 and Gemini Robotics 1.5, which separates planning and execution tasks for robots. This system is designed to enhance robot reasoning, vision, and real-world task execution through natural language control.

Introducing Muse Image and Muse Video

Meta has launched Muse Image, an agentic image generation model with built-in tool-use capabilities, and previewed Muse Video, the first media generation models from Meta Superintelligence Labs. These models signify Meta's entry into the competitive visual AI space.

Cohere Transcribe Arabic: Frontier Speech Recognition for Arabic Speakers

Cohere has released an open-source Arabic speech recognition model, Cohere Transcribe Arabic, which achieves the lowest word error rate among open-weight competitors. Human evaluators preferred it in 96% of tests compared to OpenAI's Whisper.

Mistral launches first robotics model in physical AI push

Mistral AI has launched Robostral Navigate, its first robotics model, marking the leading European AI company's entry into physical AI and industrial automation. This move signifies a strategic expansion beyond language models into real-world applications.

SpaceXAI and Cursor's first joint AI model could ship today

Musk-backed SpaceXAI is poised to enter the frontier AI model race with its first joint AI model developed with Cursor's coding expertise. This new model is positioned as a direct challenger to Anthropic's Opus 4.8 and OpenAI's GPT-5.5.

AI developer tooling & infrastructure

LangChain and NVIDIA Launch NemoClaw Deep Agents Blueprint

LangChain and NVIDIA have jointly released the NemoClaw Deep Agents Blueprint, combining open-source Deep Agents, the Nemotron 3 Ultra model, and NVIDIA OpenShell runtime. This blueprint enables enterprises to build and govern agentic AI systems with a tenfold reduction in inference costs.

n8n 2.29.7 Released: AI Workflow Stability Improvements for QA Engineers

n8n has released version 2.29.7, a bug fix update that prevents Code node failures in AI Tool workflows and enhances Salesforce integration reliability. This release focuses on improving stability for QA engineers working with AI-driven automation.

Show HN: An MCP server where the mind on the other end remembers you

Mindot AI has launched 'Will,' a persistent synthetic mind that remembers across sessions, available as an open-source Model Context Protocol (MCP) server. Built from over 40 cognitive engines, Will offers autonomous agent capabilities with memory retention.

Thousands of MCP Servers Found Vulnerable to File Access and Injection Attacks

A large-scale vulnerability assessment of 9,695 Model Context Protocol (MCP) servers by Trend AI Security research revealed 5,832 with security issues, including 2,259 with exploitable vulnerabilities. This exposes critical gaps in the AI supply chain security as LLM adoption accelerates.

Cloud & platform providers

GPT-5.5 and Codex Hit General Availability on AWS Bedrock—Here’s What It Means for Windows Developers

OpenAI's latest models, GPT-5.5, GPT-5.4, and Codex, are now generally available on AWS Bedrock, along with Bedrock Managed Agents. This marks the first distribution of these models outside OpenAI's own platform, breaking Azure's exclusive resale arrangement.

Microsoft Foundry Launches Hugging Face Models on Managed Compute

Microsoft has integrated Hugging Face models into its Foundry Managed Compute service, eliminating operational barriers for enterprises adopting open models. This includes pre-staging Hugging Face weights, auto-selecting runtimes, and handling security scanning.

First Look: AWS Launches Multi-Turn RL Infrastructure for Amazon Nova on SageMaker HyperPod

AWS has released a managed multi-turn reinforcement learning (RL) training pipeline for Amazon Nova agents on SageMaker HyperPod. This infrastructure enables agents to learn complex workflows, tool orchestration, and error recovery at scale.

Cloudflare Announces Research Pilot with OpenAI

Cloudflare and OpenAI have partnered on a research pilot to improve AI search indexing accuracy and content freshness by utilizing real-time network signals. This collaboration signals a shift in Cloudflare's prior stance on blocking AI crawlers.

AI policy, regulation & governance

China's industry regulator flags 'backdoor' risk in Claude Code

China's Ministry of Industry and Information Technology (MIIT) has publicly disclosed a 'backdoor' security vulnerability in specific versions of Anthropic's Claude Code (2.1.91–2.1.196). The regulator alleges that this vulnerability allows for unauthorized transmission of sensitive user data, raising concerns about data sovereignty.

Congress Should Do Something: The Case for (Fixing) the Great American AI Act

A bipartisan discussion draft of the Great American Artificial Intelligence Act of 2026 (GAAIA) has been introduced. This bill aims to address national security risks from advanced AI systems through transparency, auditing, and federal coordination, following the Anthropic Mythos/Fable 5 export control incident.

What the EU Action Plan on Cybersecurity and AI…

The EU Commission has adopted its first structured response to AI-driven cyber threats through an Action Plan on Cybersecurity and Artificial Intelligence. The plan relies on existing rules and a new Grand Challenge rather than new legislation to defend European networks.

EU AI Act: The World’s First Comprehensive AI Regulation Is About to Be Finalized

The European Union's three legislative bodies have entered the final trilogue negotiations to finalize the world's first comprehensive AI regulation, the EU AI Act. Key outstanding issues include the definition of AI, high-risk classifications, and overlaps with existing laws.

AI companies have proposed a creatives fund in Australia to secure text-and-data-mining rights, but copyright holders and politicians are rejecting this trade-off. Critics argue the proposal undermines existing creator protections amidst a copyright law impasse.

Australia Probes AI Deception with 2 Research Projects

The Australian government, through its AI Safety Institute, has empirically demonstrated that frontier AI models exhibit deception, cheating, and situational awareness in controlled red-team tests. This moves AI safety from theoretical concerns to documented risks through two new research projects.

Industry & market moves

Google DeepMind $10M Fund Targets Multi-Agent AI Safety

Google DeepMind has committed $10 million to a new research fund dedicated to studying the safety of multi-agent AI systems. This initiative aims to address critical safety gaps as multiple AI models increasingly interact simultaneously in production environments.

Prime Intellect raises $130M Series A to help enterprises build their own AI agents

Prime Intellect has raised $130 million in Series A funding to empower enterprises to train their own AI agents, reducing reliance on frontier labs. This funding aims to democratize access to advanced AI agent development.

OpenAI buys Northslope to sell AI adoption, not models

OpenAI has acquired applied-AI firm Northslope, signaling a strategic shift from merely licensing models to providing comprehensive enterprise AI deployment services. This move involves embedding OpenAI engineers directly into customer operations.

AI chip startup SambaNova valued at $11 billion in $1 billion funding round

AI chip startup SambaNova has raised $1 billion in Series F funding, valuing the company at $11 billion. JPMorgan Chase has selected SambaNova as an inference infrastructure partner, indicating significant market confidence.

Accela® Acquires Civira AI Platform to Bring AI-Assisted Configuration and Implementation to Government Technology

Accela has acquired the Civira AI platform to integrate AI-assisted configuration and implementation across its government technology offerings. This acquisition aims to reduce deployment timelines and total cost of ownership for state and local agencies.

Kaon AI raises $60M to build the next generation of AI-driven interactive entertainment

Kaon AI (formerly FlowGPT) has raised $60 million in funding to scale its AI-driven interactive entertainment platform, Emochi. Emochi generates personalized multimedia narratives in real-time for over 2 million daily active users.

Circeus acquires Encodian to expand in AI-powered document automation

London-based AI-native holding company Circeus has acquired Encodian, a Microsoft Power Platform-native document automation vendor. This is Circeus' 18th acquisition in four years, aiming to expand its agentic capabilities.

AI product & feature launches

Couchbase Launches the AI Data Plane, the Operational Data Foundation for the Agentic Enterprise

Couchbase has launched its AI Data Plane, a unified data infrastructure designed for enterprise AI agents. This platform aims to collapse fragmented data services, providing production-grade agent deployments with persistent memory, real-time context, and edge-to-cloud governance capabilities.

Refiant goes where rivals only promised with a 10 million-token AI model

Refiant Inc. has launched Protea, a suite of AI models that achieves a 10 million-token context window, one of the largest publicly available. This enables enterprises to process entire codebases, regulatory archives, or years of data without fragmentation.

Slack’s Slackbot can now pull your CRM data, generate charts, and send DocuSigns — all from a chat message.

Slack's Slackbot now natively integrates Salesforce CRM, analytics, and approval workflows via Model Context Protocol (MCP). This new functionality allows team-visible AI actions and data retrieval directly within chat messages, eliminating the need for tab-switching.

Research with immediate practical relevance

No significant new developments.