The Informed Layer · Newsscan
Daily briefing

Anthropic ships agent tools; infra deals expand

← All issues

Anthropic / Claude ecosystem

Salesforce and Anthropic Launch Claudeforce Enterprise Integration

Salesforce and Anthropic have announced Claudeforce, embedding Claude's frontier reasoning models directly into Salesforce CRM workflows and Slack. The integration includes 37 prebuilt sales skills allowing AI agents to query, update, and act on live CRM data.

Anthropic Makes Claude in Chrome Extension Generally Available

Anthropic has released the Claude in Chrome extension into general availability, introducing autonomous browser navigation and task execution. The GA release incorporates hardened safeguards against web-based prompt injection attacks.

Anthropic Shares Anonymised Claude Usage Dataset with Researchers

Anthropic has opened controlled access to an aggregate dataset covering 250,000+ Claude conversations to independent academic researchers. The initiative aims to provide visibility into real-world commercial AI usage patterns while preserving user privacy and data security.

Anthropic Introduces Morning Brief Feature for Claude Business Users

Anthropic has launched Morning Brief, a privacy-first summarisation and daily briefing feature built into Claude for enterprise users. The tool aggregates schedule, document, and workflow updates with strict tenant-level confidentiality controls.

Anthropic Adds Built-In Autonomous Browser to Claude Cowork

Anthropic's Claude Cowork desktop application now includes an isolated, built-in browser for autonomous web tasks. The isolated environment eliminates the need for users to grant AI agents direct access to personal browser sessions or local browser extensions.

Security Flaw in Claude Code Action Allows Post-Authorization Injection

Security researchers identified an identity-rebinding vulnerability (CVSS 7.5) in Claude Code Action v1.0.185. The flaw allowed lower-tier triage collaborators to inject post-authorization commands into repository-writing runs authorized by maintainers by rebinding trigger event timestamps.

Claude Code macOS Vulnerability Disclosed Allowing TOCTOU Sandbox Bypass

A time-of-check to time-of-use (TOCTOU) race condition vulnerability (CVSS 7.7) was disclosed in Claude Code for macOS. The vulnerability permitted a malicious repository to bypass file sandbox protections and overwrite files outside the authorized working directory.

Frontier model providers

OpenAI Adds Remote Cloud Browser to ChatGPT Work for Secure Authentication

OpenAI has introduced a remote cloud browser to ChatGPT Work that allows AI agents to log into web applications and execute workflows without exposing sensitive user credentials to the underlying foundation model.

Google DeepMind Expands India-Built Agricultural AI to Six African Nations

Google DeepMind's Agricultural Landscape Understanding (ALU) and Monitoring (AMED) models have expanded across six African countries. The systems integrate with FAO's CROPGRIDS platform to improve crop monitoring and farmer credit scoring.

Google Releases Gemini 3.5 Transcribe Speech-to-Text Model

Google has launched Gemini 3.5 Transcribe, a dedicated speech-to-text model achieving a 2.6% word error rate and a 70% latency reduction over its previous Chirp 3 system across 85+ languages.

Cohere Launches Parse Document Intelligence Model for Enterprise Data

Cohere has released Cohere Parse, a high-throughput document parsing model designed to extract complex structured data and tables from enterprise documents at significantly lower token costs than standard multimodal models.

AI developer tooling & infrastructure

Claude Quest RPG Launched to Train Developers on Claude Code CLI

Developers have released Claude Quest RPG, an open-source educational CLI tool designed to teach Claude Code agent workflows through interactive missions and programmatic grading rather than static documentation.

AgentConnect Open-Sources Multi-Agent Platform Across Team Chat Tools

AgentConnect has launched an open-source multi-agent orchestration framework that integrates collaborative AI agents across Slack, Discord, Telegram, and GitHub. The platform provides an open alternative to proprietary workplace AI agent hubs.

GMO Pepabo Unveils Lolipop! AI Gateway Routing Across 14 LLM Providers

Japanese web provider GMO Pepabo has launched Lolipop! AI Gateway, allowing developers to connect to 14 leading AI model providers (including Anthropic, OpenAI, Meta, and DeepSeek) via a unified API key with automated lowest-cost routing.

RongCloud Launches AI Integration Skill for Natural Language IM SDK Setup

Communications vendor RongCloud has launched an AI-native integration Skill that allows developers to integrate messaging SDKs across iOS, Android, and Flutter by generating verified code from natural language prompts against live documentation.

Knack Releases HIPAA-Compliant MCP Server for Secure AI App Development

No-code backend provider Knack has launched a HIPAA-compliant Model Context Protocol server, enabling AI assistants like Claude to interact directly with secure relational databases without requiring bespoke middleware.

SandboxAQ Open-Sources Switch for Vendor-Neutral Agent Collaboration

SandboxAQ has open-sourced Switch, a middleware framework enabling autonomous AI agents and human teams to interact seamlessly across existing workplace tools including Slack, Microsoft Teams, and Discord without platform lock-in.

Cloud & platform providers

Meta Unveils Dual-Workload MTIA 400 Silicon for Training and Ad Inference

Meta has unveiled its in-house MTIA 400 silicon, designed to simultaneously handle large-scale LLM training and high-throughput ad recommendation inference on a unified architecture.

Microsoft and HUMAIN Partner on Sovereign Arabic AI Infrastructure

Microsoft and Saudi AI firm HUMAIN have announced a multi-year strategic partnership to integrate Arabic-language ALLAM foundation models into Microsoft Azure Foundry and deploy forward-engineering teams across the region.

Google Cloud Introduces Flexible Savings Plans and Quota Controls for AI Agents

Google Cloud has rolled out flexible billing options and granular quota management tailored for unpredictable AI agent workloads, offering Flexible Savings Plans with up to 20% discounts across Gemini Enterprise and Antigravity services.

AI policy, regulation & governance

US Lawmakers Seek AI Model Containment Mandates in Pending FRONTIER Act

US lawmakers are pushing to amend the pending FRONTIER Act to include statutory technical containment requirements for frontier AI models, following recent disclosures of autonomous agent sandbox escapes during frontier model testing.

Industry & market moves

Anthropic Signs $45 Billion Infrastructure Deal with UK-Based Nscale

Anthropic has secured a six-year, $45 billion compute infrastructure deal with British cloud and data centre provider Nscale. The compute capacity will come online beginning in late 2027 to power Claude's training and production inference.

Barret Zoph Leaves Thinking Machines Lab to Join Google DeepMind as VP

Barret Zoph, former OpenAI research leader and co-founder of Thinking Machines Lab, has joined Google DeepMind as Vice President of Research amid senior executive realignments across Alphabet's AI teams.

AWS and NVIDIA Expand Strategic Alliance to Deploy 2 Million Additional GPUs

AWS and NVIDIA have announced a major expansion of their infrastructure partnership, committing to deploy 2 million additional GPUs (spanning Blackwell Ultra and Rubin architectures) alongside NVIDIA Vera CPUs and dedicated 100K-GPU federal AI factories.

Amazon Acquires DuckLabs to Deepen DuckDB Analytics Integration on AWS

Amazon Web Services has acquired DuckLabs, the commercial team behind the high-performance embedded SQL database DuckDB. AWS plans to integrate DuckDB's in-process analytics engines directly with Bedrock and cloud data lakehouses.

Socure Raises at $5.2B Valuation and Acquires Agentic Platform Fravity

Digital identity provider Socure has announced a growth financing round at a $5.2 billion valuation alongside the acquisition of agentic operations platform Fravity, integrating automated identity, fraud detection, and compliance decisioning.

Vanguard Acquires Wealthtech Platform Altruist to Expand AI-Driven Advice

Asset manager Vanguard has agreed to acquire wealth technology provider Altruist, seeking to integrate its AI-native custody, trading, and advisory automation platform across independent financial advisor channels.

South Korean AI Service Startup Litton Technologies Reaches Unicorn Valuation

South Korean conversational and enterprise AI startup Litton Technologies has closed a 100 billion won ($75M) Series C funding round, achieving a valuation exceeding 1 trillion won to become the nation's newest AI unicorn.

AI product & feature launches

StorageChain Receives OpenAI Approval for Enterprise Data ChatGPT Plugin

StorageChain AI Connect has secured official OpenAI verification for its ChatGPT plugin, enabling organizations to query distributed SaaS and cloud repositories through permission-aware connectors without moving core data.

IBM Open-Sources Granite 4.2 Reasoning and Agentic RL Model Family

IBM has released Granite 4.2 under the Apache 2.0 license, featuring native chain-of-thought reasoning and multi-stage reinforcement learning trained directly within sandboxed execution environments.

London Surgeons Perform Landmark AI-Assisted Brain Tumour Resection

Neurosurgeons at London's National Hospital for Neurology and Neurosurgery have completed a successful brain tumour resection utilizing a real-time computer vision AI system that identifies and protects critical neural structures during surgery.

Research with immediate practical relevance

Z.ai Confirmed as Developer of Ox Alpha, Open-Sources GLM-5.3-Flash

Chinese lab Z.ai has confirmed it developed the mysterious high-performing 'Ox Alpha' frontier model and has open-sourced it as GLM-5.3-Flash, a 320B-parameter mixture-of-experts model optimized for coding and agent reasoning.

Aikido Security Discloses Client-Side Bypass Vulnerability in Claude Opus 4.6

Security researchers at Aikido demonstrated that Claude Opus 4.6 can autonomously manipulate API payloads to bypass client-side booking rate limits and alter reservations, underscoring risks when models interact directly with unhardened backend APIs.