The Informed Layer · Newsscan
Daily briefing

Agents drive infra and policy; Anthropic accuses Alibaba

← All issues

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

Google Adds Native Computer Use to Gemini 3.5 Flash

Google's Gemini 3.5 Flash now supports native GUI control, reducing the need for bespoke API connectors and accelerating agentic automation development. This new feature enables faster and more direct interaction with computer interfaces for AI agents.

Mistral AI Connector Controls: Multi-Account Auth & Debugging | TPS

Mistral AI has added workspace-level connector controls, multi-account authentication, and a debugging tool for Model Context Protocol (MCP) connections. These features are designed to address production deployment challenges for AI agents accessing enterprise data systems.

AI developer tooling & infrastructure

Undo adds control-Z factor to AI coding agent root cause analysis - Techzine Global

Undo has launched a new AI capability that provides coding agents with deterministic runtime execution recordings. This feature aims to automatically solve hard bugs by offering deep causality data, addressing the context limitations of AI agents.

June 2026: LangChain Newsletter — Fleet On-Call Copilot, Deep Agents Rubrics, and More

LangChain's June 2026 newsletter announced several product updates, including an on-call copilot for alert triage, computer use capabilities in Fleet, and rubric-based evaluation middleware for autonomous agents.

Exclusive: LucidLink launches MCP server to give AI agents shared access to distributed files - SiliconANGLE

LucidLink has launched a Model Context Protocol (MCP) server that provides persistent, writable shared file access for multi-agent AI workflows. This enables AI agents to work from the same distributed files without requiring data replication.

The OpenRouter MCP Server — OpenRouter Blog

OpenRouter has released an MCP server that grants AI coding agents live access to real-time model pricing, benchmarks, and test inference capabilities. This allows agents to make informed decisions on model selection.

CData Launches Connect AI Developer Edition, Python SDK, and CLI • Disaster Recovery Journal

CData has released free AI developer tools, including Connect AI Developer Edition, a Python SDK, and a CLI. These tools provide governed enterprise data access through SQL, Python, and MCP interfaces to accelerate enterprise AI projects.

Jaz Open Source Agent Orchestration Framework Now Available | Tech Pulse

Jaz, an open-source agent orchestration framework, is now available, enabling persistent, long-running autonomous AI agent workflows through a client/server architecture without constant user intervention.

Cloud & platform providers

AWS GenAIIC Partner Agent Factory: New AI agents now in AWS Marketplace | AWS Marketplace

The AWS Marketplace now features seven enterprise AI agent solutions co-developed through the GenAIIC's Partner Agent Factory program. These solutions address various needs, including document translation, video intelligence, governance, customer service, clinical documentation, and financial analytics.

Inside Microsoft's 2026 AI Strategy: Unifying Azure, Copilot, and Security into a Single Control Plane - Windows News

Microsoft is planning a unified AI control plane architecture that spans Azure, Copilot, and enterprise governance. This strategy aims to address shadow AI and regulatory challenges by 2026.

Microsoft Azure Unveils Autonomous Workflow Agents: Implications for Enterprise AI Automation — Tech Daily Shot

Microsoft has launched Autonomous Workflow Agents on Azure, enabling enterprises to automate complex multi-step business processes with minimal human oversight. This offering directly competes with similar solutions from Google and Anthropic.

Microsoft Gives AI Agents an Identity—and a Leash | THE D[AI]LY BRIEF

Microsoft has released its first-party frontier and small language models (MAI) alongside a Governed Agent Stack. This governance layer provides every AI agent with a corporate identity, policy enforcement, and audit trails, removing compliance barriers to enterprise agentic AI deployment.

AI policy, regulation & governance

France's AI infrastructure investments are… · Slicast

France's AI infrastructure is transitioning from announced ambitions to operational deployment, with multiple megawatt-scale data centers and AI factories now in production. This indicates a tangible advancement in the nation's AI capabilities.

Italia steps up efforts on the AI Act: Europe’s first implementation package - Il Sole 24 ORE

Italy has become the first EU Member State to adopt a comprehensive national framework for implementing the AI Act. This framework establishes dual governance through Agid and ACN, with sector-specific oversight and strict safeguards on employment and biometric use.

RBI Draft Rules Push Kill Switches, Human Oversight for Bank AI Models

The Reserve Bank of India (RBI) has released draft guidance mandating kill-switch mechanisms and human oversight for all AI systems deployed by banks and financial institutions. It also includes enhanced validation and testing requirements for third-party models.

AI Workforce Regulation: What Employers Need to Do Now — Consumer Protection Dispatch — June 24, 2026

California and Connecticut are implementing concrete legislative requirements for employer use of AI in hiring, discipline, termination, and workforce reduction decisions. These new laws mandate human review and worker notification.

Republican Lawmaker to Introduce Bill Requiring AI Developers to Report Dangerous Incidents

A Texas Republican lawmaker plans to introduce a bill that would create a federal early-warning system. This bill requires AI developers to report dangerous capabilities, security breaches, and safety incidents to the Commerce Department within seven days, with congressional escalation for severe cases within 48 hours.

Australia is moving beyond blocking piracy to licensing copyrighted works for AI training. This involves establishing collective licensing as the official post-hoc remedy for AI companies that have already ingested creative content.

Australia Launches Agentic AI Playbook for Government

Australia's Digital Transformation Agency (DTA) has released its first comprehensive 32-page technical standard for deploying autonomous AI agents in government. This playbook establishes three oversight models (human in/on/out of loop) and lifecycle governance controls.

Whistleblower allegations suggest that the Australian government may be linking copyright carve-outs for AI training to multi-billion-dollar data center investments, contradicting official denials.

APRA warns Australia is entering a risky phase of the AI era and wants – The Aussie Corporate

The Australian Prudential Regulation Authority (APRA) has escalated its regulatory concern over the cyber risks posed by frontier AI models. It is calling for Australia's largest banks to share advanced defenses with smaller lenders to prevent system-wide vulnerability.

New Linux Foundation project aims to bring DNS-style trust to AI agents – Computerworld

The Linux Foundation has launched a new project aimed at creating a DNS-based standardized framework, the Agent Name Service (ANS), to verify AI agent identity, ownership, and trust across enterprise deployments. This addresses emerging gaps in agent-to-agent authentication and auditability.

Industry & market moves

Nvidia and AWS Forge AI Control Plane at GTC 2026 to Tame Production Workloads - Windows News

Nvidia and AWS have unified their AI control plane, integrating next-gen Vera Rubin GPUs, software-managed services, and policy-driven orchestration. This aims to reduce AI production complexity and accelerate time-to-deployment by 50%.

Amazon ups India bet with fresh $13B AI infrastructure investment | TechCrunch

Amazon has committed an additional $13 billion to expand its AI and cloud infrastructure in India through 2030. This raises its total commitments in India to $48 billion, as tech giants intensify competition for data center dominance in the emerging AI hub.

Sail Research emerges from stealth with $80 million from Kleiner Perkins | Fortune

Sail Research, founded by former Apple engineer Neil Movva, has raised $80 million in Series A funding. The company aims to build AI infrastructure for autonomous multi-task execution at scale.

TrueFoundry acquires MLOps pioneer Seldon AI to accelerate enterprise agentic AI - SiliconANGLE

TrueFoundry has acquired Seldon AI, aiming to combine its agentic AI control plane with Seldon's production-grade MLOps infrastructure for enhanced enterprise deployment.

Code Metal Acquires Signal Processing Technologies, Expanding Its Provable AI Platform and Establishing the Advanced RF Group

Code Metal has acquired Signal Processing Technologies, Inc., expanding its provable AI platform into advanced RF, signal processing, and spectrum sensing. This strengthens its capabilities for defense and commercial telecommunications.

Incode Acquires Identiq to Expand Its Privacy-First Architecture for Identity and Fraud Prevention

Incode has completed its acquisition of Identiq, with a commitment of $100 million, to integrate privacy-enhancing cryptographic technology for peer-to-peer fraud collaboration without exposing customer data.

Agility Robotics to Go Public Through $2.5 Billion Merger with Churchill Capital Corp XI | Morningstar

Agility Robotics, a humanoid robotics company with commercial deployments, is set to go public through a $2.5 billion SPAC merger with Churchill Capital Corp XI. This positions it as the first publicly listed pure-play humanoid company in the U.S.

Asteria Corporation and Pegasus Tech Ventures (USA) Launch US$10M AI and Robotics Investment Fund | Press Release | News | Asteria Corporation

Asteria Corporation and Pegasus Tech Ventures have jointly established a USD 10 million corporate venture fund. This fund will invest in physical AI and robotics startups globally over the next 10 years.

ReSource Pro Acquires Flow Specialty

ReSource Pro has acquired Flow Specialty's AI technology and team. This acquisition aims to expand its AI Factory for delivering managed agentic services in the insurance industry.

Anthropic accuses Alibaba of 'largest known distillation attack' on Claude - Nikkei Asia

Anthropic has publicly accused Alibaba of executing the largest known model distillation attack against its Claude AI model. This marks a significant escalation in allegations of AI intellectual property theft.

Scaled Cognition raises $100M led by Khosla

AI startup Scaled Cognition has raised $100 million in Series A funding led by Khosla. The company aims to build models that eliminate hallucinations through architectural redesign rather than post-hoc safety layers.

AI product & feature launches

Facebook rolls out an AI companion app for creators

Meta has launched a standalone Creator Studio app with a conversational AI assistant designed to help creators optimize content, compete with platforms like TikTok and YouTube, and reduce their reliance on third-party tools.

Zuckerberg wants Meta to launch its own prediction market, report says

Mark Zuckerberg is reportedly planning for Meta to enter the prediction market sector with a gamified, cashless app called Arena. This move aims to capitalize on the growing popularity and profits seen by competitors in the space.

Meta launches complete AI creative ad ecosystem

Meta has introduced a new AI-powered end-to-end creative ad ecosystem, providing integrated tools for affiliates and creators. This system aims to help marketers cross the AI threshold by streamlining ad creation and optimization.

AGIBOT's Spirit G2 Hits 99.98% Success Rate Across Full Tablet Inspection Line at LONGCHEER in Historic 6-Day Livestream

AGIBOT's Spirit G2 humanoid robots achieved a 99.98% success rate on a full production tablet inspection line at LONGCHEER during a 6-day livestream. This demonstrates their capability for real industrial deployment.

Watch Humanoid Robots Compete in Penalty Football from GSMA MWC26 Shanghai

Humanoid robots competed in a live penalty football demonstration at MWC26 Shanghai, showcasing advances in motion control and autonomous decision-making. This event highlights the practical capabilities of current robotic technology.

Inclusion AI Launches Ling and Ring 2.6 for Scalable Agentic Intelligence | Tech Pulse

Inclusion AI has launched its Ling and Ring 2.6 model families, introducing a dual-model architecture designed for scalable agentic intelligence. Ling-2.6 focuses on fast responses, while Ring-2.6 handles reasoning, enabling trillion-parameter intelligence on consumer hardware.

Research with immediate practical relevance

How LLM Reasoning Unlocks Hidden Factual Knowledge

Google AI researchers have discovered that chain-of-thought reasoning in large language models (LLMs) unlocks hidden factual knowledge through computational buffering and semantic priming. However, they also noted that intermediate hallucinations can degrade final accuracy.

Rough demos unlock precise robot actions, with up to fourfold real-world gains

KAIST's new research, 'DiSPo', enables robots to learn precise manipulation tasks from coarse demonstrations and autonomously adjust precision. This achieves up to fourfold real-world performance gains over existing AI models while reducing data collection costs.

Introducing Un-0: Generating Images with Coupled Oscillators - Unconventional AI

Unconventional AI introduces Un-0, the first capable image generator using a simulation of a physical dynamical system (coupled oscillators). It achieved an FID score of 6.74 on ImageNet 64x64, with open-source weights and code.

Self-Harness: AI Agents That Rewrite Their Own Rules | TechMash

Research from Shanghai Artificial Intelligence Laboratory introduces 'Self-Harness'—AI agents that automatically rewrite their own operating scaffolding (prompts, tools, runtime logic) without retraining model weights. This achieved up to 21.4 percentage point gains on Terminal-Bench 2.0.

Alibaba's model never trained as an agent — and improved agent performance across seven benchmarks - YTBlast — Tech, AI & Cybersecurity News

Alibaba's Qwen-AgentWorld models demonstrate that by training models to predict environment states rather than agent actions, they can improve agent performance across seven benchmarks without agent-specific fine-tuning.