The Informed Layer · Newsscan
Daily briefing

Anthropic agents scale enterprise; US tests models

← All issues

Anthropic / Claude ecosystem

PwC is deploying Claude to build technology, execute deals, and reinvent enterprise functions for clients

PwC is expanding its deployment of Anthropic's Claude across hundreds of thousands of professionals globally. This includes launching a new finance business group (Office of the CFO) and has already delivered production results, such as compressing insurance underwriting cycles from 10 weeks to 10 days.

Claude Code's '/goals' separates the agent that works from the one that decides it's done | VentureBeat

Anthropic has introduced a native '/goals' feature for Claude Code agents, which separates the task execution phase from the task evaluation phase. This prevents agents from prematurely exiting tasks and eliminates the need for external observability systems to manage task completion.

Claude Code Error Rate Reduced to 3% with New 12-Rule Framew | Phemex News

Anthropic has successfully reduced Claude Code's error rate to just 3% by implementing a new 12-rule behavioral framework. This framework was rigorously tested across 30 different code repositories, demonstrating significant improvements in code generation accuracy.

Claude Code Config & Pricing Updates; GPT-5.5 Codex Benchmarks & Bedrock Cost Warning

Anthropic has made 'Adaptive Thinking' the default for Claude Opus 4.6 and Sonnet 4.6, deprecating manual control over 'Extended Thinking'. This change impacts developer workflows that previously relied on precise control over reasoning parameters.

Frontier model providers

OpenAI brings Codex to mobile devices, adds more customization features - SiliconANGLE

OpenAI has expanded Codex to iOS and Android platforms, allowing developers to provide real-time guidance on long-running programming tasks without needing desktop access. The update also introduces new Hooks and Remote SSH customization features.

Introducing workspace agents in ChatGPT

OpenAI has introduced workspace agents powered by Codex that are designed to automate complex workflows and integrate with various enterprise tools such as Slack. These agents also include enterprise-grade governance controls for secure deployment.

Google DeepMind Releases Gemma 4, Its ‘Most Capable’ Open-Source AI Models – SMBtech

Google DeepMind has released the Gemma 4 family of open-source models under the Apache 2.0 licence. These models are claimed to rank third and sixth on Arena AI leaderboards, outperforming competitors up to 20 times their size.

Google pivots to Gemini Intelligence, linking AI with premium hardware

Google is redefining its Android and AI strategy around 'Gemini Intelligence,' positioning premium hardware as the primary battleground for AI innovation. This involves deep integration of Gemini across devices and partner ecosystems.

xAI unveils its first coding agent to rival Anthropic

xAI has entered the AI-assisted software development market with its first coding agent, Grok Build, which has entered early beta. This move directly targets Anthropic's established dominance in the sector.

Qwen3.6 and DeepSeek V4: China’s Open-Weight Models Now Match Frontier Competitors – ToKnow.ai

New open-weight Chinese AI models, specifically Qwen3.6-27B, DeepSeek V4-Pro, and DeepSeek V4-Flash, are now demonstrating performance on par with frontier closed-source competitors on standard benchmarks. These models also offer advantages in terms of cost and accessibility.

DeepSeek âm thầm ra mắt "cơn ác mộng thực sự" cho OpenAI: Mô hình AI mới miễn phí, chạy được ngay trên Mac Studio

DeepSeek has quietly released a 685B parameter open-weights model under an MIT license, capable of running locally on consumer hardware like the Mac Studio M3 Ultra. This move challenges OpenAI's closed proprietary model approach and aims to narrow the US-China AI capability gap.

Kimi WebBridge Turns Open Source AI Into A Local Browser Operator - Open Source For You

Moonshot AI has launched Kimi WebBridge, a local-first browser automation platform powered by its open-source Kimi models. This positions Chinese frontier AI as a direct challenger to US proprietary systems in the agent tooling market.

AI developer tooling & infrastructure

Pacvue Launches MCP Server, Making Commerce Media Data Accessible Across Enterprise AI Tools

Pacvue has launched its MCP Server, which enables enterprises to directly access commerce media data from various AI tools like ChatGPT, Copilot, Gemini, and Claude. This integration is facilitated via the open Model Context Protocol (MCP) standard.

Osaurus brings both local and cloud AI models to your Mac

Osaurus has released an open-source Mac-native LLM harness that allows users to seamlessly switch between local and cloud AI models. This system keeps files and tools on the user's hardware, addressing privacy concerns and optimizing token costs.

Device Trust MCP Server: Natural language queries for your entire fleet | 1Password

1Password has released the Device Trust MCP Server, enabling IT and security teams to query their entire device fleet data using natural language prompts directly within AI tools like Claude. This streamlines fleet management and security oversight.

Lumetra Launches Engram, an MCP-Native Memory Layer Scoring 91.6% on LongMemEval

Lumetra has launched Engram, an MCP-native memory layer designed for AI agents, achieving 91.6% accuracy on the LongMemEval benchmark. Engram offers transparent retrieval and supports bring-your-own-model integration.

Cloud & platform providers

AWS adds Advanced Prompt Optimization tool to Bedrock

Amazon Web Services (AWS) has launched the Advanced Prompt Optimization tool for Bedrock. This tool is designed to help enterprises reduce inference costs and improve the efficiency of scaling generative AI applications in production.

The AWS AI Security Framework: Securing AI with the right controls, at the right layers, at the right phases

AWS has released a structured, three-phase, three-layer security framework for AI workloads. This framework maps controls to specific use cases and deployment phases, aiming to address the governance gap where 80% of organizations adopt AI but only 10% govern it.

AWS gives Singapore students Kiro credits to build AI Skills - Techgoondu

AWS is expanding access to its Kiro AI developer tool for tertiary students in Singapore by providing 1,000 free credits. Additionally, it has launched the AWSome Lab portal to connect student AI projects with real-world enterprise challenges.

Cloudflare Introduces Workflows V2 with Deterministic Execution and 50K Concurrent Workflows

Cloudflare has introduced Workflows V2, significantly increasing its concurrent workflow capacity from 4,500 to 50,000 instances. The update also includes deterministic, replay-safe execution for distributed orchestration workloads.

Cloudflare Browser Run on Containers Is Now Faster and More Scalable - Glostarep

Cloudflare has rebuilt its Browser Run service on dedicated Containers infrastructure, resulting in a 4x increase in concurrency limits and 50% faster response times. This enhancement also enables WebGL and WebMCP support through improved state management via the D1 database.

AI policy, regulation & governance

Before the Public Sees Them, the U.S. Government Will Test Top AI Models

The U.S. government will now test frontier AI models for national security risks and other hazards before their public release. This initiative, facilitated through voluntary agreements with major AI labs, shifts AI oversight from post-launch reactivity to pre-deployment security assessment.

U.S. Government Will Test AI Models for National Security Risks, Other Hazards Prior to Release

The U.S. government is shifting from a reactive AI policy to mandatory pre-release evaluation of models for national security risks, including cybersecurity, biosecurity, and chemical weapons. This will utilize the TRAINS (Testing Risks of AI for National Security) framework and NIST CAISI benchmarks.

Australia tightens data centre scrutiny amid AI boom

The Australian Federal Government has established a formal National Interest Framework for Data Centres and AI Infrastructure. This framework sets clear expectations for projects regarding energy, water, jobs, and sovereign data objectives.

AI to power medicines approvals but humans will still call the shots

The Australian government is deploying AI to accelerate drug and housing approvals, aiming for $10.2 billion in regulatory cost savings. Human decision-making authority will be retained for final approvals in these processes.

Sam Altman’s OpenAI 'caught' sharing users data with Google, Meta; it includes email IDs and ...

OpenAI is facing a class-action lawsuit alleging that it embedded tracking pixels in its services. These pixels reportedly transmitted user conversations and personal data, including email IDs, to Meta and Google without adequate user consent.

Industry & market moves

Almost 5 months after Microsoft gave engineers access to Anthropic's Claude Code, company is canceling licenses; says: This is shared accountability to make ...

Microsoft is canceling internal licenses for Anthropic's Claude Code, five months after providing engineers with access. The company is migrating engineers to GitHub Copilot CLI by June 30, 2026, citing strategic focus and fiscal year-end cost reduction initiatives.

Cisco Job Cuts: Cisco To Cut 4,000 Jobs as AI Integration Accelerates, ETTelecom

Cisco has announced a workforce reduction of 4,000 jobs, representing less than 5% of its total workforce. These AI-driven job cuts come as the company embeds AI across its functions and increases investments in silicon, optics, and security.

FinancialContent - NTT DATA Announces Intent to Acquire WinWire to Scale Enterprise AI Adoption and Accelerate Industry Transformation with Microsoft

NTT DATA has announced its intent to acquire WinWire, which will add 1,000 Azure engineers and agentic AI capabilities to its portfolio. This acquisition aims to strengthen NTT DATA's position as Microsoft's fastest-growing GSI partner for enterprise AI transformation.

Boomi & Couchbase join forces on enterprise AI agents

Boomi and Couchbase have announced a partnership to deliver an integrated software stack for enterprise AI agents. This collaboration focuses on co-engineered data connectivity, governance, and real-time retrieval to help enterprises move AI agents from pilot to production at scale.

Accenture Federal Services And OpenAI Announce Partnership To Accelerate Secure AI Adoption

Accenture Federal Services and OpenAI have announced a partnership to accelerate secure AI adoption across U.S. federal agencies. This includes establishing an integrated implementation partnership, an Agentic Lab at The Forge, and FedRAMP-aligned implementation pathways.

Origin Lab Raises $8M in Seed Round Funding to Turn Video Game Worlds Into AI Training Data

Origin Lab has secured $8M in seed funding to commercialize licensed video game worlds as structured training data. This data will be used for AI world models and multimodal systems, enabling more realistic and complex AI simulations.

Microsoft adds more former Ai2 researchers, bolstering its Superintelligence team – GeekWire

Microsoft has significantly bolstered its Superintelligence division by recruiting at least 10 researchers from the Allen Institute for AI (Ai2), including its former CEO and core OLMo model team. This move aims to reduce Microsoft's dependence on OpenAI for frontier AI research.

FinancialContent - Experian Partners With ServiceNow to Scale Trusted Decisioning to Agentic AI

Experian and ServiceNow have partnered to integrate the Experian Ascend Platform with the ServiceNow AI Platform. This collaboration enables autonomous AI agents to access trusted data and decisioning capabilities directly within enterprise workflows, facilitating the scaling of agentic AI deployments.

AI product & feature launches

Sapphire 2026: SAP heralds dawn of ‘autonomous enterprise’ - AKEX Solutions Inc.

SAP has unveiled an integrated autonomous enterprise platform at Sapphire 2026, combining its Business AI Platform with over 50 Joule Assistants. This platform is designed to automate end-to-end business processes across finance, supply chain, and human resources.

Fiserv has co-created AI agents with six banks and OpenAI | American Banker

Fiserv, a major banking software provider, has launched agentOS, a bank-grade AI agent operating system developed in collaboration with OpenAI and AWS. This platform enables secure AI agent deployment across six bank partners, including solutions for commercial loan onboarding and report generation.

ShengShu unveils world action model to offer ‘infinite possibilities’ for robotic intelligence

ShengShu Technology has unveiled Motubrain, a world action model designed to unify perception, reasoning, prediction, generation, and action within a single embodied AI system for robotics. This model aims to replace traditional task-specific models.

Government AI chatbot goes live across GOV.UK App – PublicTechnology

The UK government's AI chatbot, powered by Anthropic's Claude LLM, has gone live across the GOV.UK App, reaching 563,000 users. It is designed to answer common citizen questions and reduce the burden on call centers.

Research with immediate practical relevance

Breakthrough Method Tackles AI Data Cannibalism | Mirage News

Researchers at King's College London have demonstrated a breakthrough method to prevent AI model collapse in closed-loop training. Their study, published in Physical Review Letters, shows that adding a single external datapoint can effectively prevent hallucinations in large language models.

TetraMem and Academic Partners Demonstrate 700°C RRAM/Memristor Breakthrough, Advancing Path Toward Deep-Space AI Computing | The AI Journal

TetraMem Inc. and academic collaborators have demonstrated RRAM (Resistive Random-Access Memory) devices capable of operating reliably at 700°C with reduced power consumption. This breakthrough in high-temperature memristors advances non-volatile memory for extreme-environment and deep-space AI computing.