The Informed Layer · Newsscan
Daily briefing

Agents and on-device AI advance; EU AI Act deal struck

← All issues

Anthropic / Claude ecosystem

Anthropic's Mythos Finds Over 10,000 Vulnerabilities in First Month of Project Glasswing

Anthropic announced that its unreleased Claude Mythos model has uncovered more than 10,000 high- and critical-severity vulnerabilities across 50 partner organizations within one month of launching Project Glasswing. The findings highlight AI's ability to discover bugs at a pace that outstrips traditional patching cycles, creating a transitional security gap.

Anthropic Broadens Access to Claude Mythos Preview's Security Tools

As part of Project Glasswing, Anthropic is expanding access to the Mythos Preview security tools for eligible customers. This move broadens the base of organizations that can use the unreleased AI model for proactive bug hunting.

Frontier model providers

Argent + Gemma 4 E4B Enable On-Device iOS Automation

A demo shows Google's Gemma 4 edge model running locally on iOS via the Argent bridge, performing end-to-end automation without cloud calls. This paves the way for private, device-resident AI agents.

OpenAI Adds ChatGPT to PowerPoint for Faster Slide Creation

OpenAI integrated ChatGPT directly into Microsoft PowerPoint, enabling AI-assisted slide creation and editing while preserving user control.

OpenAI Codex Can Now Operate a Locked Mac

Codex's Computer Use plugin now supports operation on macOS even when the system is locked, extending its autonomous capabilities.

OpenAI Launches ChatGPT Plugins Globally

OpenAI officially rolled out its plugin ecosystem worldwide, transforming ChatGPT into a connected AI assistant with real-time data access and task execution capabilities.

OpenAI Images 2.0 Drives Billion-Visual Milestone

Images 2.0 enables 2K-resolution generation with multilingual non-Latin text rendering and web-contextual reasoning, crossing a billion visual creations.

xAI Launches Grok Skills for Persistent Custom AI Workflows

xAI introduced Grok Skills, a feature that preserves custom AI behaviors across devices for selected subscribers, enabling persistent personalization.

Gemini App for Mac to Get 'Spark' AI Agent and Voice Control

Google plans to integrate its Gemini Spark agent and voice control into the Gemini macOS app this summer, expanding cross-platform AI capabilities.

Google Closed Gemini CLI to Enterprise After Accepting 6,000 Community Contributions

Google pivoted Gemini CLI from an open-source collaboration tool to an enterprise-only product with a closed-source successor, frustrating the contributor community.

AI developer tooling & infrastructure

great_cto v2.17 Adds Jurisdiction-Aware Agents and Companion Plugins

The latest version of the AI CTO tool introduces jurisdiction-aware agents, new critics, and companion plugins (Superpowers and Beads) for automated planning and task tracking.

JetBrains Positions as Last Independent AI Coding Tools Vendor

JetBrains emphasized its neutrality and governance controls via JetBrains Central, seeking to define itself as an independent alternative in the AI coding assistant market.

OrcaRouter Adds Qwen 3.7 Max Support, Targets 40% AI Agent Cost Cut

OrcaRouter integrated Alibaba's Qwen 3.7 Max model and OpenAI-compatible agent management, claiming to reduce AI agent operation costs by up to 40%.

Kore.ai Launches Artemis Enterprise AI Agent Platform on Azure

Kore.ai released Artemis, a multi-agent enterprise AI platform on Microsoft Azure, emphasizing governance and deployment flexibility.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

AI Recreated Dead Pilots' Voices, NTSB Locks Down Database

Using AI to reconstruct cockpit voices from public spectrograms led the NTSB to restrict access to its accident docket, raising concerns about data privacy and misuse.

EU AI Act Omnibus Deal: 2027 Deadlines and Narrower High-Risk Scope

The EU reached a provisional deal on the AI Act omnibus, delaying high-risk compliance to 2027, narrowing the scope, and adding a ban on non-consensual AI-generated explicit images.

Australia's AI Data Centre Boom Sparks Resource Concerns

The rapid expansion of AI data centres in Australia is prompting calls for a moratorium and clearer planning standards due to electricity and water usage fears.

ASIC Backs Australia's Fintech Edge Amid AI Transformation

Australia's financial regulator expressed support for the nation's fintech sector and emphasized balancing innovation with protections as AI reshapes financial services.

NDIS Delays Rollout of Robo-Planning System

Australia's National Disability Insurance Scheme postponed the deployment of its AI-driven robo-planning system to ensure participant safeguards and trust.

China Blocks Meta's $2B Acquisition of Manus AI

Chinese regulators blocked Meta's planned $2 billion buyout of AI agent startup Manus AI, leading the Singapore-based company to seek alternative funding at a $2 billion valuation.

Industry & market moves

PwC Rolls Out Claude Code to Hundreds of Thousands of Staff

Professional services firm PwC is deploying Anthropic's Claude Code to its global workforce, following an expanded partnership. This represents one of the largest single-organization rollouts of AI coding assistants.

Meta Drafts 7,000 Staff Into New AI Task Force

Meta is rapidly reassigning thousands of employees across the company into a centralized AI task force to accelerate its AI capabilities, signaling a strategic pivot.

Meta and Broadcom Lead $125M AI Chip Research Hub at UCLA

Meta, Broadcom, and other industry leaders are investing $125 million in a semiconductor research hub at UCLA to develop next-generation AI chips.

xAI-Anthropic Compute Deal Signals AI Infrastructure as Standalone Business

A compute capacity agreement between xAI and Anthropic highlights the emergence of AI compute as a monetizable asset, decoupling hardware from model development.

Pendo Names Zain Lakhani as First Chief AI Officer

Product experience platform Pendo appointed Zain Lakhani as its first Chief AI Officer to lead AI engineering and product observability.

Dassault Aviation Invests in Harmattan AI at €1.4B Valuation

French defence startup Harmattan AI raised €200M led by Dassault Aviation, becoming France's first defence AI unicorn at a €1.4 billion valuation.

Accenture and HUMAIN Forge AI Alliance in Saudi Arabia

Accenture and HUMAIN formed a strategic alliance to accelerate enterprise AI adoption across Saudi government and private sectors.

AI Startup Recursive Raises $650M for Self-Improving Technology

Recursive, an AI startup focused on recursive self-improvement, secured $650 million in funding, signaling strong investor confidence in rapid path to AGI.

TCS and TPG Invest ₹18,000 Crore in AI Data Centre Platform HyperVault

Tata Consultancy Services secured strategic investment from TPG to expand its HyperVault AI-ready data centre platform, with potential total funding of ₹18,000 crore.

IKS Health Acquires ARAI Solutions to Boost Agentic AI Capabilities

IKS Health acquired AI management firm ARAI Solutions to strengthen its agentic AI stack and reduce third-party dependencies in healthcare administration.

Spotify and Universal Music Sign Deal for Fan-Made AI Covers

Spotify and Universal Music Group agreed to allow paid Premium users to create AI covers and remixes of licensed tracks, with consent and compensation safeguards.

AI product & feature launches

Nigeria Launches Meta-Backed AI Chatbot for Government Information

Nigeria introduced GovGuide, an AI chatbot built on Meta's open-source models with multilingual support, to provide citizens with access to government information and services.

Zoom Expands MCP Server to Connect Conversations with Enterprise AI

Zoom enhanced its Model Context Protocol server to enable cross-tool conversation intelligence and agentic search across AI platforms and enterprise apps.

Informatica Expands AWS Integration for Enterprise AI Agents

Informatica deepened its AWS integrations to provide trusted, governed data for AI agent workloads, improving data quality and compliance.

Versa Extends Zero Trust to AI Agents and MCP Workflows

Versa introduced a patent-pending zero trust architecture for AI agents that applies identity, RBAC, and policy checks to every MCP action in production environments.

Qwen3.7-Max Runs Autonomously for 35 Hours to Optimize Chip Code

Alibaba's Qwen3.7-Max demonstrated autonomous kernel optimization for a custom chip over 35 hours, achieving a reported 10x speedup.

Thinking Machines Lab Reveals Its First Multimodal Interaction Model

Startup Thinking Machines introduced TML-Interaction-Small, a concurrent multimodal model capable of processing audio, video, and text in real time with micro-turn interactions.

China Launches SeeLight S1 Humanoid Robot for Household Chores

Chinese company GigaAI unveiled SeeLight S1, a general-purpose humanoid robot capable of cooking, mopping, and folding clothes, with deployment expected in 2027.

Research with immediate practical relevance

Google DeepMind's AlphaProof Nexus Solves 9 Erdős Problems

Google DeepMind's AlphaProof Nexus used a looped AI-proof system with formal verification to autonomously solve 9 Erdős combinatorial problems and prove 44 sequence conjectures.

Meta Paper Shows Summary Reuse Improves AI Coding Agents

A Meta research paper found that a concise two-line summary of prior interactions can outperform thousands of log tokens in guiding AI coding agents, boosting efficiency.

Nous Research Releases CNA for Sparse MLP Circuit Steering

Contrastive Neuron Attribution (CNA) identifies a tiny subset of late-layer MLP neurons that differentiate harmful from benign prompts without training or weight changes.