The Informed Layer · Newsscan
Daily briefing

Agentic AI accelerates; regulators and markets respond

← All issues

Anthropic / Claude ecosystem

Anthropic launches Project Glasswing to secure critical software against capable AI models

Project Glasswing is an Anthropic initiative to harden critical software against vulnerabilities discovered or exploited by advanced AI. An initial update outlines the project's scope and early findings.

Frontier model providers

OpenAI unveils Workspace Agents for continuous office operations

OpenAI announced Workspace Agents, a new feature set designed to handle persistent office tasks such as scheduling, document management, and workflow automation. The agents aim to replace manual office coordination with autonomous AI.

Appshots in Codex: Command-Command Captures Mac Window (2026)

A new macOS feature called Appshots allows users to attach a window screenshot with structured app text directly to a Codex conversation. This enhances context for AI-powered coding assistants.

OpenAI offers GPT-5.5-Cyber to Japan for national cybersecurity

OpenAI is expanding access to its GPT-5.5-Cyber model, tailored for cyber defense, to Japan and allied nations. The move follows heightened cyber threat activity.

Google unveils Gemini Omni, a multimodal video creation AI family

Google introduced Gemini Omni, capable of generating and editing video from text, images, and audio. The model includes embedded safeguards and avatar support for creator tools.

Google DeepMind launches inaugural AI Campus in South Korea

DeepMind established its first AI Campus in South Korea through a state-backed partnership focused on AI research, safety, and workforce development. The initiative aims to foster talent and collaboration in the region.

Meta quietly launches Forum app, a Reddit-like platform for Facebook Groups with AI-powered feed

Meta released Forum, a standalone app that transforms Facebook Groups into a knowledge-sharing platform with AI-powered group feed and admin assistant. The app is currently in limited testing.

DeepSeek makes 75% API discount permanent for V4-Pro and V4-Flash models

DeepSeek announced a permanent 75% price reduction for its V4-Pro and V4-Flash APIs, significantly undercutting rival pricing. The move aims to drive volume adoption among startups and developers.

xAI integrates Grok with OpenCode for enhanced coding

xAI has added Grok model support to the OpenCode editor, allowing Grok and X Premium subscribers to use the model directly without extra API keys. The integration aims to boost developer productivity.

Google Lighthouse experiments with agentic browsing checks using llms.txt

Google's Lighthouse performance tool added an experimental Agentic Browsing category that tests whether AI agents can navigate and complete tasks on a site. It checks for root-level llms.txt files.

Amazon Nova Act becomes HIPAA eligible, enabling autonomous browser agents in healthcare

Amazon Nova Act, a service for building browser-based AI agents, achieved HIPAA eligibility. This allows healthcare organizations to deploy agents for tasks like data entry and claim processing while maintaining compliance.

Google introduces Co-Scientist, a multi-agent AI partner for scientific research

Google launched Co-Scientist, a multi-agent system designed to structure and accelerate hypothesis generation and evaluation. The tool aims to assist researchers in formulating and testing ideas across disciplines.

AI developer tooling & infrastructure

Google launches Android CLI 1.0 for AI coding agents at I/O 2026

Google introduced a command-line interface for Android, enabling AI coding agents to interact with Android devices for testing and development. The tool is aimed at streamlining mobile development workflows.

Warp's Oz Platform can now run Claude Code and Codex alongside its own agent

Warp updated its Oz platform to act as a unified control plane for Claude Code, OpenAI Codex, and its native Warp Agent. The update adds multi-agent coordination and memory features.

Securitytrails MCP integration with Mastra AI for agentic data access

Composio released an MCP integration enabling Mastra AI agents to dynamically query Securitytrails data through a unified endpoint. The tool supports automated threat intelligence and infrastructure analysis.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

Ohio Senate passes AI regulation bill requiring watermarking and banning deepfakes

Ohio Senate Bill 163, mandating AI-generated content watermarking and deepfake prohibitions, passed the Senate and moves to the House. It represents the state's first comprehensive AI regulation.

Trump pulls AI safety executive order after tech CEO interventions

President Trump cancelled a planned executive order on voluntary AI model safety vetting following last-minute calls from Elon Musk, Mark Zuckerberg, and David Sacks. The order had aimed to coordinate developer-government interaction.

Connecticut AI bill advances with stronger compliance mandate

A Connecticut bill that strengthens AI compliance and verification requirements moved forward in the state legislature. The legislation aims to enhance accountability for AI deployments.

California Governor Newsom signs executive order to prepare for AI disruption

Governor Gavin Newsom signed a first-of-its-kind executive order aimed at shielding workers and small businesses from AI-driven job displacement. The order calls for programs to ensure equitable benefits from AI productivity gains.

South Korea finalizes enforcement decree for AI Basic Law

South Korea is completing an enforcement decree for its revised AI Basic Law, expanding protections for vulnerable groups and promoting public AI adoption. The decree supports the law's implementation.

Australian workplace department trials public ChatGPT use despite government ban

Australia's Department of Employment and Workplace Relations is running a trial allowing staff to use public ChatGPT, contrary to a broader government ban on AI tools. The trial assesses productivity benefits with safeguards.

Queensland expands government-owned AI learning platform to all state schools

Queensland is rolling out the Corella AI platform across state schools to reduce administrative workload and enhance classroom learning. The government-owned tool supports teachers with lesson planning and grading.

AI Brain Health Initiative secures $30 million in Australian government funding

The Australian Epilepsy Project received $30M to embed its AI brain health platform into the national health system, aiming to accelerate diagnosis and expand access to neurological care.

EU-Anthropic talks over Mythos AI cybersecurity concerns stall

Negotiations between the EU and Anthropic regarding testing the Mythos AI model for cybersecurity vulnerabilities have stalled, according to Spain. Limited progress was reported at a meeting of EU finance chiefs.

Industry & market moves

Anthropic projects first profitable quarter with $10.9B revenue in Q2 2026

Anthropic is reportedly on track for its first profitable quarter, driven by surging enterprise demand for Claude. The milestone comes amid a funding round that could value the company above $900B.

Cohere expands sovereign AI push with Multiverse Computing and Indra partnerships

Cohere partnered with Multiverse Computing and Indra to advance sovereign AI capabilities, focusing on secure, localized deployments for European public and private sectors. The deals complement its open-source Command A+ model.

DeepSeek founder declares AGI goal as $10 billion funding round advances

DeepSeek's founder announced an ambition to achieve artificial general intelligence while the company is raising over $10B in a new funding round. The firm prioritizes open-source research over short-term monetization.

DeepSeek targets $45 billion valuation amid AGI development push

Source: Bloomberg reports DeepSeek is aiming for a valuation around $45B in its first major external funding round, driven by investor interest in its AGI-focused open-source strategy.

Cognition/Devin acquires Windsurf (Codeium) to accelerate its coding agent ecosystem

Cognition, the company behind Devin, has acquired Windsurf (formerly Codeium), bringing its technology, IP, and staff into the Devin platform. The deal aims to expand Devin's AI coding capabilities.

AI Accelerator CRC appoints Tom Cawley as mining sector lead

Tom Cawley, chair of MaxMine, was appointed to lead the mining sector work of the AI Accelerator CRC. A production-grade ML system has been deployed across multiple mine sites, showcasing AI adoption in mining.

Accenture invests in Aera to scale AI supply chain decision-making

Accenture and Aera Technology announced a partnership to deploy AI-driven, real-time decision-making across complex global supply chains. The collaboration targets multiple industries, including manufacturing and logistics.

Symmetry Systems joins forces with Zscaler for AI-aware data security

Symmetry Systems partnered with Zscaler to enforce information flow policies across identities, permissions, data stores, and AI systems at enterprise scale. The combined solution aims to secure data used by AI workloads.

Cranium AI acquires Aiceberg to strengthen end-to-end AI security platform

Cranium AI purchased Aiceberg, uniting two AI security and governance platforms to secure the full AI lifecycle from development to autonomous agents. The acquisition expands Cranium's agentic AI security capabilities.

Coupa acquires Tonkean to accelerate agentic AI for procurement

Coupa acquired Tonkean, a no-code AI workflow platform, to integrate agentic intake and orchestration into its Coupa Compose suite. The move aims to enable fully autonomous procurement workflows.

Stellantis and Wayve partner to bring AI-powered hands-free driving by 2028

Automaker Stellantis and AI driving startup Wayve announced a partnership to integrate Wayve's AI Driver into Stellantis vehicles by 2028. The first production model is planned for North America.

AI product & feature launches

Hirundo security-hardened Gemma 4 variant outperforms LLMs 170x its size on prompt injection resistance

Hirundo released a 4B-parameter Gemma 4 variant with weight-level unlearning, achieving strong resistance to prompt-injection attacks without utility loss. It outperforms much larger models on security benchmarks.

SAP taps Mistral AI to help customers migrate legacy software

SAP is collaborating with Mistral AI to provide AI-assisted migration to S/4HANA. The offering includes automated RAG chatbot support and data-local processing in Europe, aiming to reduce migration costs and time.

Trust3 AI launches MCP Security for enterprise agentic workloads

Trust3 AI introduced MCP Security, a dedicated security and governance layer for agentic AI deployments within its Agent DOS platform. It aims to enforce policies across multi-agent ecosystems.

Nvidia's Vera CPU unlocks a new $200B market for agentic AI, says Jensen Huang

Nvidia CEO Jensen Huang claimed the Vera CPU creates a new $200 billion addressable market focused on agentic AI. The chip is designed to handle autonomous AI workloads at scale.

New tool gives personalized countdown for when AI will take your job

A website launched a tool that uses evidence-backed models to estimate how long until AI disrupts a user's specific job. The countdown is personalized based on job role and industry trends.

NVIDIA's AI agents automate signal discovery in quantitative finance

Nvidia introduced a multi-agent system based on the NeMo Agent Toolkit for automated alpha signal discovery in trading. The system handles hypothesis generation, backtesting, and refinement using Nemotron models.

Qwen introduces Qwen3.7: a reasoning agent model with 1M token context window

Alibaba's Qwen team released Qwen3.7, featuring a 1-million-token context window and a maximal reasoning mode with explicit chain-of-thought. The model targets agentic and long-context tasks.

TD Bank deploys agentic AI to slash mortgage pre-adjudication from hours to minutes

TD Bank Group launched an agentic AI model that reduces mortgage pre-adjudication time drastically. The system automates document review and eligibility checks, aiming to improve customer experience.

Anker Innovations unveils THUS AI chip platform for compute-in-memory local AI

Anker introduced the THUS AI chip, a compute-in-memory design that enables local AI processing in consumer devices like earbuds. The platform promises lower power consumption and faster inference.

Research with immediate practical relevance

DeepSeek releases paper on V3 model cost reduction method

A new paper authored by DeepSeek's Liang Wenfeng details a cost-reduction method for the V3 model, addressing hardware bottlenecks in training and inference. The technique is published for public review.

Cloudflare finds inconsistent AI safety refusals in Glasswing security testing

In testing frontier AI models for cybersecurity research, Cloudflare's Project Glasswing uncovered inconsistent safety refusals that could undermine reliability. The findings call for improved governance and reliability measures.