The Informed Layer · Newsscan
Daily briefing

Agents proliferate; guardrails and charters try to keep up

← All issues

Anthropic / Claude ecosystem

Barbars Scales Claude Across Operations and Client Experience

Barclays is expanding its use of Anthropic's Claude to improve customer service and operational efficiency while maintaining governance and security controls. The deployment marks a major regulated-bank scaling of Claude in the UK.

Anthropic Launches Claude Frontier Academy Aiming to Train 10,000 AI Engineers

Anthropic is expanding its credentialing and training ecosystem with a dedicated Claude Frontier Academy, targeting 10,000 engineers by 2027. Executive Steve Corfield detailed the programme in an interview.

Anthropic Makes Claude for Government Generally Available to Agencies

Claude for Government is now generally available to US federal and state agencies with no seat fees and fixed usage caps under FedRAMP High authorization. This extends Anthropic's government footprint to civilian agencies.

Claude Code Deletes User Files via NTFS Junction Bug on Windows

A newly documented bug in Claude Code on Windows wiped user folders including Documents and Downloads due to an NTFS junction edge case combined with agent command execution. The incident highlights destructive-command risks in coding agents.

Frontier model providers

Grok Bot Expands Coding Tools with Cursor, GitHub, and Origin Integrations

xAI's Grok Bot is expanding its toolset with Cursor integration, GitHub/Origin PR management, and video demos, broadening its software-development automation capabilities.

Google Introduces Gemini 4 Argon with 1M Token Output Limit

Google officially introduced Gemini 4 Argon, its next frontier model with a 1M token output limit aimed at scalable, memory-efficient, code-heavy enterprise workflows. A phased, guardrailed rollout to government and trusted defenders is underway.

Meta Launches Ray-Ban Meta Gen 3 AI Glasses in India with Muse AI Features

Meta announced Ray-Ban Meta Gen 3 in India with longer battery life, a slimmer design, and new AI features, alongside Ray-Ban Meta Audio and Muse AI agent availability. Ranveer Singh was named brand ambassador for the launch.

Meta Muse Crosses 5 Million Downloads in 22 Days with 3 Million Weekly Users

Meta's Muse AI agent has surpassed 5 million downloads in 22 days, with weekly active users topping 3 million. The figures signal Muse's shift from conversational AI to an action-executing agent across Meta apps and services.

Cohere Releases Embed 5 Pro and Embed 5 Fast Embedding Models

Cohere launched a split embedding system: Embed 5 Pro for indexing and Embed 5 Fast for querying, operating in a shared vector space. The design can reduce index sizes while preserving retrieval quality.

AI developer tooling & infrastructure

DeepKeep Launches AI Lens for Developer Coding Agent Security

DeepKeep introduced AI Lens for Developers, a lightweight guardrail and monitoring plug-in giving CISOs visibility into AI coding agents and AI-assisted workflows. It flags data leaks, routes destructive commands for approval, and provides full auditability.

GitHub Copilot Gains Computer Use to Interact with Desktop Apps

GitHub Copilot can now autonomously interact with desktop applications via a computer-use capability in public preview, extending automation beyond code to GUI-only software.

C1.ai Launches C1 LLM Gateway to Govern Enterprise AI Model Routing

C1.ai launched the C1 LLM Gateway, providing governed routing for enterprise AI traffic with policy-based model access and per-call cost attribution.

AI Coding Agents Leak 13,000 Internal Images to GitHub

A newly reported incident shows AI coding agents exfiltrated and publicly exposed roughly 13,000 internal images via GitHub repositories. The report includes recommended controls to prevent agent-driven data leaks.

DeepSeek Harness v0.2 Launches Desktop Version Ready to Use Out of the Box

The DeepSeek Harness team released v0.2 with first-time desktop installations and enhanced plugin management, with about 60% of users employing third-party plugins. The coding agent is now accessible to general office users.

JetBrains Opens EAP for Air, an Agentic-First IDE Experience

JetBrains opened the Early Access Program for Air in IDEs, introducing an agentic-first UX enabling parallel agent work alongside existing agents with IDE-integrated controls.

Kong Ships AI Gateway 2.2 with MCP Tool Governance and Identity Policies

Kong released AI Gateway 2.2 to general availability, adding MCP tool governance, identity policies, and expanded model/provider support for strategic portability in enterprise AI deployments.

Cloud & platform providers

OpenAI GPT-6 Astra Supports UltraFast Mode on Amazon Bedrock

AWS announced UltraFast mode for OpenAI's GPT-6 Astra on Amazon Bedrock, enabling up to 6x faster inference for latency-sensitive workloads.

AWS Debuts Strands Decider 2B, a Lightweight Decision Model for Agentic Workflows

AWS introduced Strands Decider 2B, a token-efficient lightweight decision model designed to accelerate agentic workflows by avoiding heavy text generation and reducing latency in agent decision points.

Azure Container Apps Express Reaches GA on Newly Available Sandbox Layer

Microsoft brought Container Apps Express to general availability on a new Container Apps Sandbox layer featuring subsecond startup, per-second billing, scale-to-zero, hardware-isolated microVMs, and snapshot resume.

AWS Launches Gamified Agentic AI Training Experiences for Developer Upskilling

AWS launched three gamified, live-sandbox AWS Jam training experiences covering its agentic AI portfolio to bridge the developer skills gap in enterprise production environments.

Cloudflare Introduces Managed Cloudflare OS as a Company Agent Workspace

Cloudflare announced Cloudflare OS, a fully managed, customisable agent workspace for organisations, offered via a waitlist for managed deployments.

Cloudflare Introduces Clef Open-Source Decision Models and RL Fine-Tuning Platform

Cloudflare released Clef, a set of open-source decision models, alongside a new reinforcement-learning fine-tuning platform.

AWS Demonstrates Kiro Sample Agents Cutting SAP Modernization Effort by Up to 87%

AWS published a workflow using Kiro sample agents for SAP modernization, claiming up to 87% reduction in ABAP modernization effort across ABAP, BW, and PI/PO workloads.

AI policy, regulation & governance

Rep. Jayapal Introduces National Charter System Framework to Rein in AI

Rep. Pramila Jayapal introduced a legislative framework establishing a national charter system for AI, enabling broad government oversight and accountability before harms occur.

Australia Directs Agencies to Strengthen Cybersecurity After OpenAI Hack

Australia issued a directive requiring non-corporate Commonwealth entities to strengthen cybersecurity against AI-enabled threats and legacy tech risks, with specific remediation deadlines.

Australia Prepares New AI Rules to Protect Creative Content

Australia is moving to regulate AI training data use to protect creative content, with a framework that could require consent and compensation from technology companies training on Australian works.

Sens. Hawley and Murphy Push AI Liability Legislation as Trump Backs Self-Regulation

Sens. Josh Hawley and Chris Murphy are advancing the AI Agent Accountability Act, a bipartisan liability framework for AI, in contrast to the White House's preference for self-regulation.

California Governor Newsom Signs First-in-the-Nation AI Worker Protections

Governor Newsom signed SB 947, SB 951, AB 1331, and AB 1883, adding first-in-the-nation AI worker protections covering AI in workplaces and healthcare with transparency and privacy safeguards, strengthening California's AI framework.

Australian Unions Propose Big Tech Levy to Fund Australian-Owned AI

Australian unions and AI researchers proposed a levy on large technology data centres to pool government funding and compute for developing an Australian-owned AI capability.

Industry & market moves

Kodiak AI Chooses AWS for AI-Powered Autonomous Trucking Development

Kodiak AI selected AWS to scale development of its autonomous trucking AI, including the Kodiak Driver and BreakPoint AI-powered safety validation tool, ahead of a driverless highway service launch.

Bloomberg Completes Canoe Intelligence Acquisition for AI-Powered Private Markets Workflow

Bloomberg completed its acquisition of Canoe Intelligence, integrating Canoe's private markets data extraction capabilities into a unified, AI-powered end-to-end private and public market investing workflow.

TCN Acquires Abstrakt to Expand AI Agent Assist and Automated QA

Contact center platform TCN acquired Abstrakt to integrate real-time AI agent assist and automated QA capabilities into its AI-powered contact center offerings.

Nebius Acquires Stealth AI Startup Inferize in $100-150M Deal

Nebius acquired 10-month-old stealth startup Inferize for a reported $100-150 million, expanding its AI infrastructure stack to reduce idle GPU costs and scale demand-driven inference.

Analysys Mason Acquires AI/ML Consultancy Modulai

Analysys Mason acquired Swedish AI and machine learning consultancy Modulai, expanding its specialist AI capabilities and giving Modulai scale across Analysys Mason's client base.

AI Hardware Design Startup Flow Engineering Raises $50M at $750M Valuation

Flow Engineering, an AI hardware design tool provider, raised a $50 million Series B at a $750M valuation backed by Valor, Atreides, and Sequoia, signalling venture momentum in AI-assisted chip design tooling.

Action-Model AI General Intuition Raises $220M at $6.2B Valuation

General Intuition, a startup focused on action-models for inside-world actions, raised $220 million at a $6.2 billion valuation, betting on models that act within simulated and real environments.

AI product & feature launches

Blockworks Makes Crypto Data Searchable from ChatGPT, Claude, and Other AI Tools

Blockworks launched an MCP data server enabling AI tools to query its crypto market data covering 40,000+ assets with sources attached, directly from ChatGPT, Claude, and other AI clients.

Perplexity Releases pplx-embed-v2-context-9b-preview Contextual Embedding Model

Perplexity released pplx-embed-v2-context-9b-preview, a contextual embedding model that retrieves answers along with their supporting evidence, trained to emphasise verifiable context over a single gold passage.

Decagon Introduces Voice 3 Agent with Chord Speech Model

Decagon unveiled Voice 3, its most advanced voice AI agent, paired with a bespoke Chord speech model and four related AI concierge products for customer service automation.

Research with immediate practical relevance

UW-Meta Paper Proposes Language Models That Edit Their Own Context

A new University of Washington-Meta paper introduces Context Language Models that can edit and manage their own context as a file. The approach could simplify agent-harness engineering by shifting prompt orchestration into model weights.

Google Reports AI Finding More Medium-Risk Security Flaws

Google reported that AI is rapidly expanding the set of known vulnerabilities but tends to identify medium-risk flaws rather than high-risk ones. The finding reshapes how defenders should prioritise AI-assisted vulnerability discovery.

AI 'Mind-Reading' Tool Reconstructs What You're Looking At from Brain Scans

Weizmann Institute researchers unveiled a high-resolution brain-decoding AI that reconstructs viewed images from fMRI data and can predict brain activity from stimuli, raising notable ethical and privacy concerns.

AI Scientist Eve Designs, Runs, and Learns from Lab Experiments Autonomously

Researchers at Chalmers University of Technology demonstrated the Eve AI laboratory system autonomously designing experiments, running them in a self-driving lab, and learning from results — AI progressing from tool to active scientific agent.

Topos Bio Releases Topos-2 and Topos-Bind Protein AI Models

Topos Bio released Topos-2, setting a new state of the art on PeptoneBench, and Topos-Bind, advancing all-atom ensemble generation for protein-ligand complexes including an amyloid-β case study.

NVIDIA Researchers Introduce Physis-Lang, Lifting Cosmos 3 Past Veo 3.1 on Physics Benchmarks

NVIDIA researchers introduced Physis-Lang, a self-evolving physics-grounded captioning language that improves physics reasoning in video understanding and pushes Cosmos 3 past Veo 3.1 on multiple benchmarks, with fully local execution possible.