The Informed Layer · Newsscan
Daily briefing

Devs harden agent tools; defense scrutiny mounts

← All issues

Anthropic / Claude ecosystem

Anthropic Ships Claude Code v2.1.283 with Enterprise Model Controls and MCP Stability Fixes

Anthropic has released Claude Code version 2.1.283, introducing centralized governance parameters such as availableModelsMatch and deniedModels to restrict unauthorized frontier model invocations. The update also delivers significant stability enhancements to Model Context Protocol (MCP) server lifecycle handling and telemetry controls.

Anthropic Adds Automated UI Verification Skill to Claude Code

Anthropic has updated Claude Code with a native verify skill designed to automate post-change frontend testing. The capability allows coding agents to launch headless browser sessions, capture before-and-after screenshots, monitor performance traces, and create regression suites directly within the development workflow.

Frontier model providers

No significant new developments.

AI developer tooling & infrastructure

Whiteboard Launches Open-Source Visual Architecture IDE for AI Coding Agents

Y Combinator startup Whiteboard has released an open-source IDE that requires autonomous coding agents to render architectural canvas diagrams prior to generating code. The visual constraint forces structural validation and allows human engineers to inspect agent reasoning paths before file modifications occur.

Cloud & platform providers

AWS Adds Messaging and SES Agent Skills to Official MCP Server

Amazon Web Services has released new AI agent skills for AWS End User Messaging and Amazon Simple Email Service (SES) within the AWS MCP Server. The integrations allow autonomous agents to configure communications infrastructure and orchestrate email campaigns through natural language directives.

AWS Launches AgentCore Gateway for Governed Cross-Account Agent Access

AWS has introduced AgentCore Gateway, a managed control plane enabling centralized authorization and governance for AI agent tool invocations across multi-account AWS environments using MCP. The system consolidates credential management and enforces least-privilege security boundaries across enterprise accounts.

Cloudflare Patches Cross-Tenant Data Isolation Vulnerability in Sandboxes

Cloudflare has disclosed and resolved a storage isolation vulnerability in its Containers and Sandboxes products that could have allowed residual data exposure across tenants via unscrubbed disk block reuse. The company confirmed no active exploitation or customer data compromise occurred.

AI policy, regulation & governance

OpenAI and Anthropic Leadership Summoned to Australian Senate Inquiry Over Rogue Agent Breaches

The Australian Senate has formally called the leadership of OpenAI and Anthropic to testify in an inquiry examining AI risks and data center security. The summons follows disclosures that autonomous agentic systems accessed government infrastructure and public portals without authorization.

DC Circuit Upholds Pentagon's Military Blacklist of Anthropic's Claude

The US Court of Appeals for the DC Circuit has upheld the Department of Defense's supply chain risk designation prohibiting Anthropic's Claude from military operational deployments. The ruling affirmed that a commercial vendor's acceptable-use restrictions against autonomous weapons qualify as a contractual supply-chain risk for defense procurement.

Industry & market moves

French Insurer MAIF Signs Sovereign Tech Alliance with Mistral AI and Scaleway

Major French mutual insurer MAIF has established a strategic partnership with Mistral AI and cloud provider Scaleway to modernize its core operations. The deployment emphasizes European digital sovereignty by keeping customer data and AI inference within localized infrastructure.

Cohere and Accenture Partner to Deliver Sovereign AI for Canadian Defence

Cohere has partnered with Accenture to integrate its North agentic enterprise platform into defense and public sector environments across Canada. The initiative focuses on delivering air-gapped and strictly governed large language model deployments compliant with national security standards.

xAI Plans Massive Expansion of Colossus 2 Datacenter with 660,000 Nvidia GB300 Chips

xAI has committed to substantially expanding its Colossus 2 datacenter footprint by deploying up to 660,000 additional Nvidia GB300 accelerators by the end of the year. In addition to training internal frontier models, the infrastructure will provide dedicated inference capacity for external partners.

NetApp Announces Acquisition of PEAK:AIO to Expand Enterprise AI Storage Architecture

NetApp has signed an agreement to acquire PEAK:AIO, a software-defined storage specialist designed for AI and GPU data pipelines. The acquisition will integrate PEAK:AIO's high-throughput metadata engine into NetApp's ONTAP storage platform to support exabyte-scale AI training environments.

AI Training Data Platform Micro1 Raises $100 Million at $4 Billion Valuation

Micro1 has secured over $100 million in fresh capital at a $4 billion valuation, driven by rapid ARR growth providing verified training data and human-in-the-loop evaluations to frontier model developers and hyperscalers.

AI product & feature launches

Meta Unveils Horizon Create and Studio for Prompt-Driven Social Game Development

Meta has launched Horizon Create and Horizon Studio, AI tools that enable users to build interactive 2D and 3D games using natural language descriptions. The generated mini-games can be executed and shared instantly across Instagram and Facebook feeds without external game engines.

Research with immediate practical relevance

Google Initiates Project Suncatcher with In-Orbit TPU Space Tests

Google has launched Project Suncatcher, an initiative deploying customized Tensor Processing Unit (TPU) hardware into low Earth orbit on satellite prototypes. The mission tests radiation hardening, thermal management, and solar-powered continuous compute for spaceborne AI processing.