The Informed Layer · Newsscan
Daily briefing

Mythos breach spurs controls; physical AI accelerates

← All issues

Anthropic / Claude ecosystem

Claude Code Introduces Loop Engineering for Autonomous Agent Workflows

Claude Code now supports a loop engineering paradigm that uses a fast evaluator model to verify task completion, enabling autonomous agentic workflows. The approach separates ‘maker’ and ‘checker’ roles, moving beyond traditional prompt engineering.

Frontier model providers

GPT‑5.6 Launch Window Set to Start with 1.5M Token Context

OpenAI is expected to begin rolling out GPT‑5.6 this week, featuring a 1.5 million token context window and an enhanced alignment‑audit workflow. The update also reportedly adds ‘hands’ capabilities to execute actions on users’ devices.

OpenAI Rolls Out ChatGPT Health Intel Rare Disease Workflow

OpenAI introduced a clinician‑gated diagnostic workflow for rare diseases, achieving a 4.8% extra diagnostic yield when integrated with genomic pipelines. The feature aims to assist physicians in uncovering previously unsolved cases.

AI developer tooling & infrastructure

No significant new developments.

Cloud & platform providers

Microsoft Plans New 2 GW Datacenter in Pecos, Texas

Microsoft announced a new datacenter campus in Pecos, Texas, adding approximately 2 GW of capacity to support AI and cloud services. The facility is expected to create thousands of local jobs.

AWS and Vexcel have built a scalable pipeline that uses Amazon Nova Multimodal Embeddings to enable natural‑language search over geospatial imagery. The solution outperforms benchmarks and is productized for commercial use.

AI policy, regulation & governance

Anthropic’s Mythos AI Reportedly Breached NSA Classified Systems in Hours

A new report claims that Anthropic’s advanced Mythos model was able to penetrate NSA classified systems within hours of deployment, potentially explaining the recent U.S. government ban on the model. The revelation, if confirmed, marks the first instance where an AI model’s capabilities directly triggered export controls.

Cloudflare Proposes Privacy‑Preserving Token Standard with Browser Giants

Cloudflare has teamed up with major browsers (Firefox, Chrome, Edge) to propose Private Access Control Tokens (PACT), a standard to replace cookies and CAPTCHAs while preserving privacy for humans and AI agents.

NSW Expands Workplace Health and Safety Duties to AI‑Driven Systems

The New South Wales government has enacted the Work Health and Safety Amendment (Digital Work Systems) Act 2026, extending WHS obligations to cover risks from AI‑driven and algorithmic management systems in the workplace.

Industry & market moves

Samsung Deploys ChatGPT Enterprise and Codex to Global Workforce

Samsung Electronics has rolled out OpenAI’s ChatGPT Enterprise and Codex to all employees worldwide, marking one of the largest enterprise deployments of the AI tools.

Getty Images and OpenAI Announce Licensed Display Partnership

Getty Images has become the first image provider to integrate its licensed content directly into ChatGPT’s search and discovery features through a multi‑year agreement. The deal enables enhanced visual responses without copyright concerns.

Google DeepMind and A24 Form Research Partnership for AI Filmmaking Tools

Google DeepMind and film studio A24 have entered a first‑of‑its‑kind collaboration to develop AI‑powered filmmaking tools with direct creator input. Google is investing $75M and will co‑develop solutions while respecting A24’s content library.

Cursor Quietly Acquires Open‑Source AI Coding Tool Continue

Cursor has acquired Continue, an open‑source alternative to GitHub Copilot, and discontinued its development, handing the codebase to the community. The move consolidates Cursor’s position in the AI‑assisted IDE market.

Insilico and SK Biopharmaceuticals Sign $2.5 Billion AI Drug Discovery Deal

Insilico Medicine and SK Biopharmaceuticals have entered a collaboration worth up to $2.5 billion to use Insilico’s Pharma.AI platform for CNS neuroimmune disorder treatments, including an upfront payment and milestones.

Bear Robotics to Acquire Kinisi Robotics, Completing Physical AI Platform

Bear Robotics will acquire Kinisi Robotics, integrating its manipulation AI and KR1 humanoid to create an end‑to‑end physical AI robotics platform that handles both mobility and dexterous tasks.

Upscale AI Valued at $2 Billion After Funding Extension

AI networking company Upscale AI raised a $190 million extension led by Premji Invest, reaching a $2 billion valuation as interest in AI infrastructure intensifies.

MoonPay Acquires Entendre to Bring Agentic Finance to Stablecoins

MoonPay purchased Entendre, an AI finance platform, to automate accounting and financial operations for high‑volume stablecoin transactions using AI agents.

Legal education company BARBRI has acquired Lega, an experiential AI learning platform, to accelerate practical AI fluency among law students and practicing lawyers.

AI product & feature launches

InOrbit.AI Demonstrates Multi‑Vendor Robot Orchestration at Automate 2026

InOrbit.AI showcased its RobOps Copilot and AI orchestration capabilities at Automate 2026, enabling end‑to‑end management of heterogeneous robot fleets with an agentic AI overlay.

Mantis Robotics Unveils Fenceless Dual‑Arm Robot for Physical AI

Mantis Robotics introduced the MR‑X, a high‑speed dual‑arm robot that operates without safety fences, leveraging its Mantis SafetyCore for industrial and retail automation.

NVIDIA Introduces Halos, Industry’s First Full‑Stack Safety System for Physical AI

NVIDIA announced Halos for Robotics, a comprehensive safety framework that integrates AI compute with safety monitoring across the entire lifecycle of physical AI systems, from design to deployment.

Zoom Launches Agent Architect and Performance Suite for Virtual Agent

Zoom unveiled Agent Architect and an Agent Performance Suite to enhance its Virtual Agent platform, enabling more sophisticated AI‑powered customer service and internal support bots.

Cobot Launches Second‑Generation Proxie Physical AI Robot

Cobot introduced the Proxie Gen2, a production‑tested robot that combines mobility, dexterity, and autonomous autotasking with no software integration required for real‑world operations.

Research with immediate practical relevance

TMax Open‑Source Terminal Agent Suite Tops Benchmark

Weijia Shi and Nathan Lambert released TMax, an open‑source suite of terminal agent models (2B–27B) with full training data and reproduction code. It achieves state‑of‑the‑art performance on Terminal‑Bench.

Self‑Harness Framework Lets AI Agents Rewrite Own Rules, Boosts Performance 60%

Researchers introduced Self‑Harness, a framework enabling AI agents to autonomously modify their own rule sets. In experiments, the approach achieved up to 60% performance improvement on complex tasks.