The Informed Layer · Newsscan
Daily briefing

Agents drive innovation; Regulation, security grow

← All issues

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

OpenAI launches an AI-powered documentation assistant on its developer portal to generate custom integration guides

OpenAI launched an AI-powered documentation assistant on its developer portal. This new tool generates custom integration guides and allows for export as Markdown, making developer documentation more conversational and interactive.

OpenAI launches Codex browser developer mode with Chrome DevTools Protocol integration for 2x faster automation

OpenAI's Codex launched a new Developer Mode with Chrome DevTools Protocol integration. This allows for deeper debugging access, achieving 2x faster browser automation by inspecting runtime errors, DOM state, and network traffic.

Lyria 3: Google DeepMind’s High-Fidelity Sonic Revolution | AI Magazine

Google DeepMind's Lyria 3 achieves professional-grade, long-form music generation up to three minutes with structural coherence. The model is now integrated into the consumer-facing Gemini app and features SynthID safety watermarking.

AI developer tooling & infrastructure

Introducing Browserless Agent: The Fastest Browser Agent

Browserless Agent is introduced as the first MCP-native browser agent. It combines persistent sessions, command batching, and a stateful architecture to outperform competitors across six real-world benchmarks.

Atropos Health unveils Evidence Agent MCP on Databricks

Atropos Health launched its Evidence Agent MCP on Databricks Marketplace. This new offering aims to bring trusted real-world evidence directly into AI agents for clinical decision-making, enhancing the accuracy and reliability of healthcare AI.

Introducing Omnigent: A Meta-Harness to Combine, Control and Share Your Agents | Databricks Blog

Databricks launches Omnigent, an open-source meta-harness that unifies multiple AI agent harnesses (Claude Code, Codex, Pi) under a single interface. It features collaborative, policy-driven, and multi-model composition capabilities.

Cloud & platform providers

Workers Changelog | Cloudflare Docs

Cloudflare introduces Sandboxes, a secure container-based environment for Workers to execute untrusted AI-generated code on-demand. This enhances security by isolating potentially malicious code.

AI policy, regulation & governance

Anthropic disables access to Fable 5, Mythos 5 on government directive

The U.S. government issued an export control directive forcing Anthropic to immediately disable access to its Fable 5 and Mythos 5 models to all foreign nationals, citing national security concerns. This marks a significant escalation in government control over frontier AI deployments following a reported jailbreak discovery.

Tech executives to attend G7 summit as leaders address AI, online safety - CNA

Tech executives from leading AI companies have been invited to attend the G7 summit to discuss crucial topics such as AI regulation, AI infrastructure, and online safety protections. The summit aims to engage industry leaders directly in shaping global policy for emerging technologies.

Canada’s move to regulate AI chatbots after Tumbler Ridge shooting raises doubts over loopholes - The Globe and Mail

Canada introduces legislation to regulate AI chatbots and restrict social media for under-16s following the Tumbler Ridge shooting. However, experts are warning of potential enforcement loopholes and a lengthy implementation timeline for the bill.

Australia has valuable cards to play in the AI future, but critical gaps remain, new report finds - ABC News

The Tech Policy Design Institute has released its AI Agency Tool, an evidence-based framework designed to assess Australia's AI capabilities across 103 dimensions. This tool aims to inform policymaking and shift national discourse from sovereignty to strategic agency.

AI music in Australia: Paul Dempsey and Gordi call for mandatory labelling of AI-generated tracks on streaming services

Australian musicians Paul Dempsey and Gordi are advocating for mandatory AI-generated music labelling on streaming platforms. This call comes amid growing concerns that a significant portion—up to 44%—of uploads involve no human creativity.

Industry & market moves

Mistral AI seeks to raise €3B at €20B valuation as Europe's AI race heats up

Mistral AI is seeking to raise €3 billion in funding at a €20 billion valuation, nearly doubling its previous valuation. This move signals Europe's intensifying competitive AI race and the convergence of semiconductor and AI supply chains, attracting investors like ASML.

DeepSeek史诗级宕机13小时，一夜崩上热搜，网页版更新，V4真来了？-36氪

DeepSeek experienced its longest service outage to date, lasting 12 hours and 58 minutes. This incident sparked widespread speculation that a new V4 model, potentially with improved coding and multimodal capabilities, might be undergoing a silent rollout.

NTT DATA boosts Microsoft AI capabilities with WinWire buy | TechMarketView

NTT DATA has acquired Microsoft-specialist partner WinWire to strengthen its Azure AI delivery capabilities. This move aims to enhance NTT DATA's competitive position against major players like Accenture, Capgemini, and Infosys in the enterprise AI transformation market.

AI product & feature launches

Grok creator Igor Babushkin introduces personalized River AI project – Zamin.uz, 12.06.2026

Igor Babushkin, co-architect of Grok, has launched River AI, a new personalized AI agent system. River AI adapts to individual user preferences and aims to serve as a persistent digital companion rather than a replacement.

Physical Intelligence Built One AI Model That Can Run Different Robot Bodies Without Retraining

Physical Intelligence has developed a generalist robot model that can operate on various hardware platforms without task-specific retraining. This model, available in pi0 and pi0.5 Vision-Language-Action versions, combines perception, planning, and motor control into a single neural network.

Now, AI agents can decide when to strike deals using UPI mandate - The Times of India

Pine Labs launched India's first agent-to-agent payment capability, the Pine Labs Payment Protocol (P3P). This enables AI agents to autonomously execute financial transactions through an extended UPI mandate framework without requiring human authentication at the point of transaction.

Research with immediate practical relevance

Google DeepMind Maps the Road From AGI to Superintelligence: Four Paths and Hard Limits

Google DeepMind researchers, including AIXI framework creator Marcus Hutter and DeepMind co-founder Shane Legg, have published a report mapping four technological pathways from artificial general intelligence to superintelligence. They argue that this transition will be gradual rather than a dramatic step change and will be bounded by fundamental physical limits.

New AI model called "Count Anything" does exactly what it says, and that's harder than it sounds

Tsinghua University researchers have developed "Count Anything," a single AI model that reliably counts objects across diverse image types, including satellite imagery and medical scans, using only text prompts. It consistently outperforms specialized competitors.

Technion tool lets users create AI videos by mouse

Researchers at the Technion-Israel Institute of Technology have developed a plug-in technology called 'Time to Move (TTM)' that enables users to control motion in AI-generated videos via simple mouse gestures. This innovation works without requiring model retraining or massive computing resources.

Prompt injection breaks today’s AI agents, study warns | CSO Online

The StakeBench research consortium's study reveals that current AI web agents lack dependable defenses against prompt injection, with attack success rates ranging from 41.67% to 79%. The study also identifies multi-stakeholder harms, including stealthy parasitism where attackers succeed undetected.