The Informed Layer · Newsscan
Daily briefing

Policy and legal fights intensify; AI agents boom

← All issues

Anthropic / Claude ecosystem

Claude Code adds Safe Mode and fallback model chains

Claude Code introduced a three-step fallback model chain and a safe-mode isolation flag to diagnose configuration-caused session issues, reducing pipeline outages. The feature was documented in a developer guide.

Frontier model providers

DeepMind opens scientific AI tools to Asia-Pacific startups via new accelerator

Google DeepMind launched a SuperAI Singapore accelerator program giving Asia-Pacific startups access to its scientific AI toolkit for tackling environmental challenges, expanding regional AI research capacity.

Meta Edits desktop app to add AI assistant for content ideas

Meta is expanding its Edits app to desktop and integrating a new AI assistant that can brainstorm content ideas and analyze insights, with beta features rolling out.

DeepSeek introduces OCR feature to compress long text processing

DeepSeek released a new OCR feature that converts text into visual tokens for optical compression, significantly reducing compute requirements and improving reading efficiency for long documents.

Gemma 4 open-weight models now available on Amazon Bedrock

Google's Gemma 4 family of open-weight, multi-architecture models with built-in reasoning and function calling launched on Amazon Bedrock across multiple regions.

Moonshot AI launches Kimi Work, a local desktop agent with 300-sub-agent swarm

Moonshot AI released Kimi Work, a downloadable desktop AI agent running on Kimi K2.6 that features local file access, live browsing, and a 300-sub-agent swarm architecture for knowledge workers.

Google rolls out AI Mode information agents to Ultra subscribers

Google began rolling out AI Mode information agents in Search to Ultra subscribers, enabling continuous background monitoring, web-linked updates, and multi-language support across markets.

AI developer tooling & infrastructure

JFrog brings supply chain security to Claude Code via new plugin

JFrog launched a plugin for Claude Code that introduces enterprise-grade software supply chain governance, enabling end-to-end traceability and security controls within AI-assisted development.

Xiaomi open-sources MiMo Code AI coding agent, claims outperformance on long-horizon tasks

Xiaomi released MiMo Code under an MIT license, a coding agent with a memory-management architecture designed for complex 200-step software tasks. The agent reportedly beat Claude Code in over 1,200 test scenarios.

Concentrate AI launches LLM gateway with free enterprise controls

Concentrate AI announced an LLM gateway that provides free, built-in enterprise control features and unified access to multiple AI models via a single API integration.

LangChain framework hit with critical CVEs exposing sensitive data

Several critical vulnerabilities were disclosed in LangChain core and LangGraph checkpointing, potentially exposing files, secrets, and conversation histories in AI applications.

Databricks Marketplace now offers ready-to-use MCP servers for healthcare agents

Databricks added MCP servers from multiple partners to its Marketplace, enabling healthcare organizations to rapidly build AI agents with pre-built integrations, governed within Databricks environment.

Cloud & platform providers

EC2 Capacity Blocks for ML now available in AWS GovCloud regions

AWS made EC2 Capacity Blocks for ML generally available in GovCloud US regions, allowing government and regulated customers to reserve GPU capacity for training and inference with cross-account sharing.

AWS launches autonomous FinOps agent for AI cloud cost governance

AWS introduced a FinOps agent that autonomously monitors and routes AI-related cloud cost alerts, providing real-time governance to manage spiraling compute expenses in AI projects.

AWS Graviton5-powered M9g and M9gd instances reach general availability

AWS announced general availability of EC2 M9g and M9gd instances powered by Graviton5 processors, optimized for AI-centric workloads such as inference and machine learning training.

AI policy, regulation & governance

Anthropic proposes $200M funding to study AI's impact on jobs and economy

Anthropic proposed a $200 million commitment to research AI's effects on the workforce and broader economy, signaling industry-led measures to cushion disruption. The proposal was made in the context of a broader policy push that includes funding and regulatory asks.

Arizona advances multiple AI bills, one signed into law

Arizona’s 2026 legislative session saw progress on four AI-related bills; HB 2592 was signed, while three others (HB 2133, HB 2311, SB 1786) await final votes, covering transparency, deepfakes, and algorithmic accountability.

Canada proposes AI chatbot regulation bill after school shooting incident

In response to a school shooting linked to AI chatbot interactions, Canada drafted a bill that would create a digital regulator and ban social media for under-16s, though critics warn of loopholes.

Ireland targets AI Omnibus simplification during EU presidency

As it assumes the rotating EU Council presidency, Ireland plans to prioritize the Digital Omnibus on AI, aiming to simplify and potentially delay certain AI Act obligations while focusing on child safety online.

Australian government agencies fail first AI transparency disclosure test

A significant number of Australian government agencies missed the initial deadline for mandatory AI transparency disclosures, revealing widespread non-compliance with the policy meant to ensure public accountability in AI use.

Western Australia to establish public sector AI Centre of Excellence with $10M fund

The WA government announced a new Public Sector AI Centre of Excellence backed by a $10 million investment fund to accelerate AI initiatives across state agencies.

Perplexity AI vs Amazon: Ninth Circuit case could decide legality of AI agents on the web

Perplexity AI is fighting Amazon in a Ninth Circuit case that will determine whether the Computer Fraud and Abuse Act (CFAA) can be used to block AI agents from accessing websites, potentially setting a global precedent.

Industry & market moves

xAI fires engineer who raised safety alarms about Grok; lawsuit claims retaliation

A former xAI engineer filed a lawsuit alleging he was fired after repeatedly warning about inadequate security in the Grok model. The case surfaces amid SpaceX's IPO timeline, raising concerns about internal safety governance.

Equal AI raises $30M Series B for AI call assistant platform

Equal AI, an AI-powered call assistant startup, raised $30 million in a Series B round led by Prosus and Tomales Bay Capital to expand its platform in India.

Vertiv acquires ThermoKey to bolster AI data center cooling portfolio

Vertiv completed the acquisition of ThermoKey, expanding its heat rejection solutions specifically for AI data centers, with enhanced manufacturing capabilities in EMEA.

Relativity acquires Gavel for AI document automation and contract review

Legal tech firm Relativity acquired startup Gavel, integrating its document automation and contract review capabilities into the RelativityOne platform via Microsoft Word integration.

THEKER raises $85M, Europe's largest robotics Series A, attracting Samsung and LVMH

Barcelona-based robotics startup THEKER raised an $85 million Series A from CRV, Samsung, and LVMH, marking Europe's largest round in the sector and featuring existing deployment in Inditex facilities.

Minerva raises $20M funding, collaborates with OpenAI

AI startup Minerva secured $20 million in funding from The General Partnership and others, and also announced a collaboration with OpenAI, though details remain undisclosed.

Artus AI raises pre-seed to build AI-native product management platform

India-founded Artus AI closed a pre-seed round to scale its platform that uses AI to help product teams define and build the right software, targeting global expansion.

Google sues cybercrime ring that turned Gemini into phishing tool

Google filed a lawsuit against a cybercrime group that abused Gemini’s API to create phishing campaigns, marking one of the first major legal actions against AI-powered cybercrime linked to a frontier model.

AI product & feature launches

Zyphra releases Zamba2-VL hybrid Mamba2–Transformer vision-language models

Zyphra launched Zamba2-VL, a family of open vision-language models that use a hybrid Mamba2–Transformer architecture, cutting time-to-first-token by about an order of magnitude compared to conventional transformers.

India launches sovereign Varya AI video model, open for self-hosting

India’s MeitY, in partnership with Avataar AI, launched Varya, a culture-focused AI video generation model that is faster and cheaper than global alternatives and available for self-hosting and fine-tuning.

Pine Labs introduces AI agent payments on India's UPI via payment protocol

Pine Labs launched a payment protocol (P3P) that enables AI agents to conduct UPI transactions with upfront consumer mandates, allowing automated, condition-based payments in India.

Research with immediate practical relevance

Meta FAIR unveils five breakthroughs advancing human-like AI

Meta's Fundamental AI Research team announced five research breakthroughs aimed at enhancing human-like AI capabilities, covering areas such as reasoning and planning.

Tian Yuandong's startup Recursive achieves SOTA on NVIDIA GPU kernel optimization leaderboard

Recursive, a startup founded by AI researcher Tian Yuandong, reached state-of-the-art on NVIDIA's official SOL-ExecBench benchmarks for GPU kernel optimization, marking its first public milestone.

AI accelerates molecular simulations 10,000-fold, study finds

A study published in Science Advances demonstrates an AI model that can predict molecular evolution 10,000 times faster than traditional methods, potentially revolutionizing drug discovery and materials science.