The Informed Layer · Newsscan
Daily briefing

AI agents everywhere; billions flow, rules tighten

← All issues

Anthropic / Claude ecosystem

Anthropic Ships 60+ Claude Code Reliability Fixes and Research Updates

Anthropic delivered over 60 reliability fixes for Claude Code, alongside research on agentic misalignment training, sandbagging mitigation, model spec midtraining, and Natural Language Autoencoders. These improvements enhance Claude Code’s stability and safety for developers.

Anthropic Launches Public Bug Bounty Program on HackerOne

Anthropic initiated a bug bounty program on HackerOne, inviting external security researchers to find and report vulnerabilities in its systems. The program is part of its broader Mythos security initiative.

Frontier model providers

OpenAI Adds Codex Chrome Extension for Automated Signed-In Web Tasks

OpenAI introduced a Codex Chrome extension that operates inside signed-in browser sessions, allowing it to automate enterprise web tasks with approval gates and tab-group isolation. This brings Codex’s capabilities directly into the browser for workflow automation.

OpenAI Opens GPT-5.5-Cyber Model to Vetted Cybersecurity Researchers

OpenAI is granting access to GPT-5.5-Cyber, a specialized model tuned for cyber defense, to vetted security researchers. The model includes tighter accountability and identity controls, aimed at helping identify software vulnerabilities and monitor deployments.

Google Meet Adds End-to-End Live Speech Translation on Mobile

Google Meet now offers live speech translation on mobile devices, providing real-time dubbed audio in a single language pair per meeting. The feature builds on existing captioning to make cross-language meetings more seamless.

Meta Unveils Hatch: Agentic AI Assistant for Everyday Life

Meta introduced Hatch, a new agentic AI assistant with multimodal reasoning designed to autonomously perform tasks across Meta’s ecosystem, including Instagram. Hatch is expected to replace traditional search and discovery methods with personalized, proactive AI interactions.

Meta Outlines Expanded AI Glasses Range with New Features

Meta detailed its AI glasses lineup, including Ray-Ban Meta, Ray-Ban Display, Oakley Meta Vanguard, and Oakley Meta HSTN. Sales are rising as the glasses gain new AI-powered features, enhancing everyday utility.

Meta Adds AI Agentic Workflow to Immersive Web SDK

Meta integrated AI agentic workflows into its Immersive Web SDK, allowing developers to automate coding, testing, and validation for WebXR experiences. The goal is to reduce friction and accelerate development of immersive web apps.

DeepSeek Adds Image Recognition Capability, Expanding Beyond Text

DeepSeek officially entered the era of image-text interaction by launching an image recognition mode, enabling the model to understand and process images. This expands the model’s utility for multimodal tasks.

AI developer tooling & infrastructure

Deepclaude Cuts Claude Code-Like Experience Costs by 94% Using Cheaper Backends

An open-source tool named Deepclaude allows developers to maintain a Claude Code-like autonomous coding agent experience while routing calls to cheaper backends such as DeepSeek V4 Pro via OpenRouter, reducing model costs to one-seventeenth. This dramatically lowers the barrier for teams wanting agentic coding without the premium price.

Windsurf 2.0 Redefines IDE as Mission Control for Multi-Agent Development

Windsurf 2.0 introduces Kanban-style agent management and Spaces, positioning the IDE as a central control hub for a multi-agent workforce. It competes directly with VS Code and Cursor by offering a more agentic development experience.

GitHub Copilot to Deprecate Grok Code Fast 1 Model on May 15

GitHub Copilot announced it will deprecate the Grok Code Fast 1 model across all experiences effective May 15, 2026. The move reflects ongoing model curation in the Copilot platform.

JetBrains ReSharper 2026.2 EAP Opens AI Agent Ecosystem for Visual Studio

JetBrains announced the ReSharper 2026.2 Early Access Program, integrating an open AI agent ecosystem via ACP and Junie. This allows agent-controlled AI operations within Visual Studio, moving beyond simple code completion.

Sakana AI’s RL Conductor Orchestrates GPT-5 and Claude Sonnet 4 with 7B Model

Sakana AI released RL Conductor, a 7B-parameter RL-trained model that dynamically orchestrates frontier LLMs like GPT-5 and Claude Sonnet 4. It reduces API calls while improving performance on reasoning and coding benchmarks.

AnySearch Launches as Search Infrastructure Built Specifically for AI Agents

AnySearch launched a search infrastructure platform designed for AI agents, offering unified API access to high-quality, structured data across multiple verticals. It positions itself as essential infrastructure for AI-era execution.

Cloud & platform providers

AWS Launches Claude Platform: Anthropic’s Native Platform Now Available Through AWS Accounts

AWS becomes the first cloud provider to offer direct access to Anthropic’s Claude Platform via customer AWS accounts, with billing through AWS Marketplace. This streamlines enterprise procurement and integration of Claude’s AI capabilities. The general availability follows earlier previews and expands AWS’s AI ecosystem.

Google Cloud Launches Gemini Enterprise Agent Platform

Google Cloud introduced the Gemini Enterprise Agent Platform, integrating model selection, building, deployment, governance, and ecosystem tools into one platform for enterprise automation. It aims to simplify creating and managing AI agents at scale.

Amazon Bedrock Adds Cross-Account Guardrails for Centralized AI Safety Governance

AWS introduced cross-account safeguards for Amazon Bedrock Guardrails, allowing centralized management of AI safety policies across multiple accounts. This reduces manual overhead and ensures consistent governance for generative AI applications.

AI policy, regulation & governance

Florida Governor Signs Law Regulating Large-Scale AI Data Centers

Florida Governor Ron DeSantis signed SB 484 into law, establishing regulations for large-scale AI data centers. The law aims to prevent ratepayers from bearing transmission and generation costs, ensuring that AI infrastructure growth does not shift costs to consumers.

APRA Warns Australian Financial Institutions on AI Risk Governance Gaps

The Australian Prudential Regulation Authority (APRA) issued a warning that governance and technical controls are lagging behind AI adoption in the financial sector. It signaled stronger supervisory action to ensure regulated entities manage AI risks appropriately.

Industry & market moves

Lyrie.ai Joins First Batch of Anthropic’s Cyber Verification Program

Lyrie.ai, a security layer for autonomous AI agents, joins Anthropic’s cyber verification initiative. The company also released an open Agent Trust Protocol (ATP) for real-time agent verification. This program aims to enhance trust in AI agent operations.

OpenAI Launches $4B Enterprise AI Deployment Company, Acquires Tomoro

OpenAI is creating a new $4 billion deployment company, backed by TPG, Advent, Bain, and Brookfield, to embed frontier AI engineers inside large enterprises. It also acquired AI consultancy Tomoro to add 150 deployment specialists, accelerating practical AI adoption.

DeepMind Alum David Silver Raises $1.1B for Pure Reinforcement Learning AI Startup

Ineffable Intelligence, founded by DeepMind’s David Silver, raised a record $1.1 billion seed round. The startup aims to build AI that learns purely through trial and error, without human training data, targeting capabilities that exceed current models.

Broadcom and Meta Announce Multi-Year Custom AI Silicon Partnership

Broadcom and Meta have entered a multi-year agreement to develop and scale custom Meta Training and Inference Accelerator (MTIA) chips, with Broadcom providing silicon, networking, and packaging through 2029. The partnership aims to enhance Meta’s AI infrastructure efficiency.

Meta and NVIDIA Announce Long-Term AI Infrastructure Partnership

Meta and NVIDIA have entered a multi-year collaboration, where NVIDIA will supply computing and networking technologies—including Spectrum-X Ethernet and the Vera Rubin platform—across Meta’s data centers. The deal also includes confidential computing technology to secure AI workloads.

xAI Partners with Cursor in $10B AI Coding Deal with $60B Acquisition Option

xAI is investing $10 billion in AI coding startup Cursor, with an option to acquire it for $60 billion in 2026. The deal aims to supercharge Cursor’s Composer product and leverage xAI’s compute resources for training.

Elon Musk Plans $55B Terafab Chip Factory in Texas for SpaceX, xAI, and Tesla

Elon Musk is planning a $55 billion chip fabrication plant, called Terafab, in East Texas. The facility would supply advanced microchips for SpaceX, xAI, and Tesla under one roof, aiming to secure chip supply for Musk’s AI and aerospace ventures.

DeepSeek Valued at $50B in State-Backed Funding Round

DeepSeek is raising a funding round led by a Chinese state-backed fund, valuing the company at $50 billion—up from $45 billion earlier talks. The investment underscores government support for China’s AI large-model sector as a ‘national fate battle’.

Google Cloud Commits $750M to Accelerate Partners’ Agentic AI Development

Google Cloud announced a new $750 million partner fund to accelerate adoption and development of agentic AI solutions across its ecosystem. The funding aims to support system integrators, ISVs, and startups building on Google Cloud’s AI platform.

ELECTRA AI to Go Public via SPAC Merger with Iron Horse Acquisition II

ELECTRA AI and Iron Horse Acquisition II Corp. (Nasdaq: IRHO) announced a business combination agreement valued at over $250 million. Upon closing, the combined entity will operate as ELECTRA AI, ticker AIBR, bringing a new AI company to public markets.

Nscale Secures $790M to Expand AI Data Center in Norway

Nscale raised $790 million in additional financing to support the buildout of its AI data center in Narvik, Norway, adding 115MW of capacity. This marks the largest AI infrastructure investment in Norway to date.

German AI Robotics Firm Sereact Raises $110M Series B

Sereact, a German AI robotics company, raised $110 million in Series B funding to scale its Cortex 2.0 platform and enter the U.S. market. The platform enables robots to learn and adapt in unstructured environments.

DNV Ventures Invests in Juniper Ventures Fund II for AI Assurance

DNV Ventures announced an investment in Juniper Ventures Fund II, which focuses on AI assurance technologies. The fund’s first close was announced alongside initial investments, signaling growing interest in AI trust and safety startups.

Korean Manufacturing Giants Back Robot Data Platform Config

Config, a platform dubbed the ‘TSMC of robot data,’ received seed funding from Korea’s largest manufacturers. The startup provides standardized, high-quality data to train AI robotics, lowering the barrier for robot development.

Brookfield Commits $500M to OpenAI Deployment Partnership

Brookfield Asset Management is investing $500 million in a strategic partnership with OpenAI’s Deployment Company. The investment aims to scale enterprise AI deployment infrastructure, with Brookfield also providing real estate and energy assets.

BBVA Joins OpenAI’s New Deployment Company to Accelerate Enterprise AI Transformation

Spanish bank BBVA is one of the first enterprises to sign on with OpenAI’s newly formed Deployment Company, aiming to accelerate its AI transformation. The partnership will embed frontier AI engineers within BBVA to build customized solutions.

AI product & feature launches

Writer Launches Autonomous AI Agents for Enterprise Workflow Automation

Writer announced autonomous AI agents that can operate across multiple enterprise apps to automate complex workflows without human intervention. The platform aims to handle end-to-end processes, pushing further into lights-out automation.

China’s X-Humanoid Wise KaiWu Agent Brings Real Awareness to Robots

X-Humanoid launched the Wise KaiWu agent, claimed to be the industry’s first global scene perception and dynamic memory system for embodied intelligence. The agent has been validated on real robots, giving them improved real-world interaction capabilities.

Red Hat AI 3.4 Targets Enterprise AI Deployment at Scale

Red Hat released version 3.4 of its AI platform, adding support for enterprise-scale inference, agent management, and hybrid-cloud operation with built-in governance and security features. The update targets production AI workloads across on-premises and cloud environments.

Alation Launches AI Governance Offering to Close the Oversight Gap

Alation introduced an AI Governance product that provides a single system of record for AI assets with live, board-ready compliance posture. The tool aims to help enterprises govern their AI initiatives as adoption outpaces governance.

AI-Powered Google Finance Expands to Europe with Local Language Support

Google Finance is rolling out its AI-powered features to European markets, including local language support. The expansion follows a U.S. launch and brings smarter financial insights and analysis to European users.

Research with immediate practical relevance

Prompt Injection Attacks Now Targeting Live AI Agents in the Wild

Independent security disclosures reveal that attackers are exploiting indirect prompt injection vulnerabilities to steal runtime credentials from AI agents on live websites. The attacks bypass existing defenses and highlight the growing security risks of agentic AI.

Sakana AI and NVIDIA Release TwELL for 20% Faster LLM Inference and Training

Sakana AI and NVIDIA introduced TwELL, a set of sparse CUDA kernels that achieve 20.5% inference speedup and 21.9% training speedup for large language models by exploiting activation sparsity. The technology is compatible with existing GPU hardware.

Researchers Develop AI Training Method That Reduces Overconfidence

South Korean researchers created a brain-inspired pre-training warm-up technique that allows AI models to recognize when they don’t know an answer, reducing overconfidence and hallucinations. The model can output ‘I don’t know’ instead of generating plausible-sounding but incorrect responses.