The Informed Layer · Newsscan
Daily briefing

Anthropic doubles down on agents; Regulation heats up

← All issues

Anthropic / Claude ecosystem

Anthropic Expands Claude for Legal With New Plug-Ins - Technology Org

Anthropic has expanded its Claude for Legal product with new plug-ins covering six practice areas and Model Context Protocol (MCP) connectors, integrating with major legal tools like Docusign, Box, and Thomson Reuters Westlaw.

Anthropic Traces Six Weeks of Claude Code Quality Complaints to Three Overlapping Product Changes - InfoQ

Anthropic conducted a post-mortem, tracing six weeks of Claude Code quality complaints (March–April 2026) to three overlapping product-layer changes—a reasoning effort downgrade, a caching bug, and a system prompt verbosity limit—all of which are now resolved. The company identified gaps in internal testing that allowed these silent quality regressions.

Anthropic's Claude subscriptions no longer include Agent SDK and claude -p usage

Anthropic is separating Agent SDK and claude-p usage from standard subscription interactive credits, introducing a dedicated monthly programmatic credit pool effective June 15th. This change shifts users to metered billing once their tiered programmatic credits (e.g., $20 Pro, $100 Max 5x, $200 Max 20x) are exhausted, ending previous subsidized allowances.

Claude Code Increases Weekly Limits by 50% Through July 13, 2026

Anthropic is temporarily increasing Claude Code's weekly usage limits by 50% through July 13, 2026. This move is a direct response to competitive pressure from OpenAI's Codex, which reportedly consumes significantly fewer tokens for comparable coding tasks.

Anthropic亲自下场，又一批Agent创业公司死掉了-36氪

Anthropic has launched a fully managed agent platform, 'Claude Managed Agents,' which consolidates the infrastructure previously built by numerous startups. This move effectively eliminates the need for an independent middleware layer for agent orchestration and execution.

Frontier model providers

Google Launches AI-Powered Magic Pointer for Chrome Desktop Users - The Technology Express

Google's Magic Pointer, powered by Gemini, enables context-aware cursor interactions in Chrome, allowing users to interact with AI assistance through gestures rather than typing, thereby reducing friction.

‘Gemini Spark’ is Google’s upcoming AI agent in the Gemini app

Google is developing 'Gemini Spark,' an advanced AI agent feature for the Gemini app. It is designed to learn user preferences and autonomously accomplish tasks such as email management and meeting briefings.

Mistral Developing New AI Model for Banks Lacking Mythos Access

Mistral AI is reportedly developing a new cybersecurity-focused AI model to compete with Anthropic's Mythos, specifically targeting European banks that currently lack access to Anthropic's limited-availability tool.

Google’s Gemini AI now understands your mouse cursor movements

DeepMind's Gemini-powered AI cursor now understands mouse movements and voice commands, allowing users to interact with AI through gestures rather than lengthy text prompts.

Google rolling out AI-powered ‘Contextual suggestions’ on Android: Here’s what it is

Google is rolling out an AI-powered 'Contextual suggestions' feature on Pixel 10/10a devices running Android 16. This on-device feature learns from user habits and location to deliver personalized recommendations.

AI developer tooling & infrastructure

Building a safe, effective sandbox to enable Codex on Windows

OpenAI engineers developed a custom Windows sandbox for Codex, their AI coding assistant, to enforce safe file write and network restrictions. This new implementation achieves safety parity with existing macOS and Linux versions without requiring administrative elevation.

ReSharper 2026.2 Agent Client Protocol: How JetBrains Just Cracked Open Visual Studio's AI Monopoly - Sean Kim — Arts and Tech

JetBrains has introduced the Agent Client Protocol (ACP) in its ReSharper 2026.2 Early Access Program, an open protocol designed to allow any AI agent to integrate seamlessly with Visual Studio, thus challenging GitHub Copilot's dominance in the IDE.

Coding Agent Index: Cursor Tops First Cross-Stack Benchmark | aiHola

Artificial Analysis has published the first cross-stack coding agent benchmark, which compares complete harness-and-model combinations rather than just underlying language models. The benchmark revealed that Cursor's harness significantly outperforms Anthropic's own Claude Code wrapper when both use the Claude Opus 4.7 model.

Introducing Semaphore for AI Agents: An AI-Native Developer Experience for CI/CD - Semaphore

Semaphore has launched 'Semaphore for AI Agents,' an AI-native CLI and agentic interface that enables AI coding assistants to interact with CI/CD pipelines using natural language and structured machine-readable outputs.

Cloud & platform providers

Microsoft turns Copilot Studio into an AI agent control center - Help Net Security

Microsoft has expanded Copilot Studio into a comprehensive AI agent control center with April 2026 updates, introducing centralized governance, security posture tracking, workflow automation with AI reasoning, and integration of GPT-5.5 Reasoning across Microsoft 365.

Microsoft Is Retiring Copilot Mode On Edge, Because Everything Is Copilot Mode Now

Microsoft is consolidating Copilot functionality directly into the Edge browser across desktop and mobile platforms, effectively retiring the standalone Copilot Mode feature.

AWS Enables AI Agents to Drive WorkSpaces Desktops | Let's Data Science

AWS is now enabling AI agents to automate tasks on WorkSpaces virtual desktops by exposing a managed Model Context Protocol (MCP) endpoint. This allows agents to interact with traditional desktop applications without the need for legacy application modernization.

AWS Expands AI Security with Full Code Scanning Preview

AWS is expanding its AI security offerings with a new full repository code scanning feature in preview, embedding autonomous AI agents into the software development lifecycle. This enables comprehensive, repository-wide security scanning with architectural context and automated remediation advice.

AI policy, regulation & governance

Hackers claim sale of Mistral AI source code after supply chain attack

Hackers are claiming to have stolen and are offering to sell Mistral AI's internal source code repositories. This alleged breach follows a supply chain attack that compromised SDK packages on npm and PyPI registries.

Joint White Paper Plots Sustainable Path for AI, Cloudflare, and the Future of Independent Journalism — Open Markets Institute

The Open Markets Institute, Center for Journalism and Liberty, and Washington Monthly Institute have released their first joint white paper, examining how AI exacerbates power imbalances between technology platforms and journalism, with infrastructure providers like Cloudflare acting as gatekeepers.

EU AI Act “Omnibus” Update: More Time to Prepare – A Practical Perspective | SGS USA

The EU has extended compliance deadlines for high-risk AI systems under the AI Act to December 2027 and August 2028, reducing overlap with sectoral legislation while harmonized standards mature.

The Irish Human Rights and Equality Commission (IHREC) has raised concerns regarding the independence, resourcing, and transparency mechanisms of Ireland's proposed General Scheme of the Regulation of Artificial Intelligence Bill 2026 during a committee appearance.

AI Omnibus: German government approves compromise • Table.Briefings

The German government has officially approved the EU AI Omnibus compromise reached in trilogue negotiations, signaling strong support for the impending formal adoption of the comprehensive AI regulation.

AI education guidelines updated by European Commission | Digital Watch Observatory

The European Commission has updated its AI ethics guidance for schools to address the implications of generative AI, student data protection, and the use of high-risk systems, aligning with the EU AI Act.

Personal Information Protection Act Amendment Allowing Use of Personal Data for AI Training Passes Political Affairs Committee - The Asia Business Daily

South Korea's Political Affairs Committee has passed an amendment to the Personal Information Protection Act, allowing the use of personal data for AI training under government oversight. This move aims to position the nation to develop 'sovereign AI' capabilities.

Australia AI identity governance lags as risks rise

Australian organizations are lagging behind global peers in AI identity governance, deploying AI agents for security tasks faster than they can establish adequate control frameworks.

Australian Organisations Racing To Deploy AI Agents Without Adequate Identity Security Controls, Semperis Study Finds – SMBtech

A new study by Semperis reveals that Australian organizations are deploying AI agents with access to critical identity infrastructure at a faster pace than they are implementing security controls. Only 21% of these organizations expressed confidence in their ability to recover from a credential exposure.

Asia Pacific governments boost sovereign AI priorities

Sovereign AI has risen from seventh to second in investment priorities for Asia Pacific governments, with 46.1% evaluating and 36.1% actively piloting such initiatives. This shift is primarily driven by concerns over geopolitical resilience and data control.

Industry & market moves

Anthropic forms $200 million partnership with the Gates Foundation

Anthropic has entered into a $200 million partnership with the Gates Foundation, committing grants and Claude credits to support programs focused on global health, education, and economic mobility in low- and middle-income countries.

PwC and Anthropic expand alliance for enterprise agentic AI

PwC and Anthropic are expanding their strategic alliance to deploy agentic AI, including Claude Code and Cowork, across PwC's global professional network. This initiative has already led to delivery cycle compressions of up to 70% and is targeting $2 trillion in enterprise technical debt.

新浪人工智能热点小时报丨2026年05月14日00时_今日实时人工智能热点速递|广东省|ChatGPT|Meta AI|OpenAI|AI_新浪新闻

All 11 co-founders of xAI have reportedly departed the company, including core technical talent such as pre-training lead Zhuang Juntang, following organizational turmoil.

Microsoft Eyes AI Startup Acquisitions to Reduce Reliance on OpenAI, ETEnterpriseai

Microsoft is reportedly exploring multiple acquisitions of AI startups, including Cursor and Inception, as a strategy to reduce its dependence on the OpenAI partnership and accelerate independent AI model development.

Parallel Web Systems Lève 100 M$ à 2 Md$ [2026]

Parallel Web Systems has successfully raised $100 million in a Series B funding round, valuing the company at $2 billion. This triples its valuation in five months, driven by the increasing demand for AI infrastructure to facilitate agent web access.

BMS and Hengrui Strike a $15.2B Partnership Spanning 13 Programs • Isomorphic Labs Raises $2.1B for AI Drug Discovery - BioMed Nexus

Bristol Myers Squibb and Jiangsu Hengrui Pharmaceuticals have formed a $15.2 billion partnership across 13 programs, marking the largest China-to-West licensing deal in biopharma history. Simultaneously, Isomorphic Labs has secured $2.1 billion in funding, the largest AI drug discovery round ever.

Carta Launches Carta Law with Acquisition of Avantia

Carta has acquired Avantia to launch Carta Law, which is positioned as the largest AI-native integrated legal and compliance solution for private markets. This acquisition aims to unify legal workflows with fund operations on a single platform.

Pivotal acquires LendingScore to accelerate AI technology strategy - The Intermediary - Latest UK mortgage news

Pivotal has acquired LendingScore to integrate AI-driven mortgage CRM, sourcing technology, and automated income verification into its new Innovations Lab, accelerating its AI technology strategy.

Akamai Technologies: $205 Million Acquisition Of LayerX Advances AI Workforce Security Strategy

Akamai Technologies is acquiring LayerX for $205 million to strengthen its Zero Trust security portfolio. LayerX provides browser-layer AI usage control and governance capabilities.

Boomi announces new partnerships, acquisition - BusinessWorld

Boomi has acquired Lunar.dev, an innovator in AI gateways, to extend its Model Context Protocol (MCP) gateway and governance capabilities for enterprise AI agent management.

Recursive: $650 Million Raised At $4.65 Billion Valuation For Self-Improving AI Research

Recursive, a startup founded by former AI leaders from OpenAI, DeepMind, and Meta, has raised $650 million in funding at a $4.65 billion valuation. The capital will be used to develop self-improving AI systems capable of automated scientific discovery.

OpenAI preparing ‘legal action’ against Apple over Siri partnership: report - 9to5Mac

OpenAI is reportedly preparing legal action against Apple, alleging a breach of contract related to their Siri-ChatGPT integration partnership. The claims stem from subscription revenue falling far short of projections.

Department of War Selects Applied Intuition to Build Enterprise Autonomy Pipeline

The Chief Digital and AI Office (CDAO) has selected Applied Intuition to build a unified Autonomy Factory platform. This platform is intended to accelerate autonomous system development and deployment across all U.S. military branches.

AI product & feature launches

Tencent Cloud to End Calls for 3 DeepSeek Models on May 22

Tencent Cloud is set to discontinue API access for three specific DeepSeek model versions—DeepSeek-V3-0324, DeepSeek-V3.1-Terminus, and DeepSeek-R1-0528—requiring developers to migrate their workloads before May 22, 2026.

ServiceNow and Google Cloud introduce new enterprise AI solutions and agents - techpartner.news

ServiceNow and Google Cloud have integrated their AI platforms via open protocols (Agent-to-Agent, MCP) to enable autonomous agents across 5G, retail, and IT operations. These agents are designed to autonomously detect and resolve issues.

Microsoft, Palo Alto Networks Find Many Vulnerabilities by Using AI on Their Own Code - SecurityWeek

Microsoft and Palo Alto Networks have reported significant vulnerabilities discovered using frontier AI models for internal code scanning. This indicates that AI-driven vulnerability discovery is becoming a practical security tool.

8x8 AI Studio Adds OpenAI's GPT Realtime 2 to Support Production Voice Agents – Company Announcement - FT.com

8x8 AI Studio now integrates OpenAI's GPT Realtime 2, designed to improve voice agent reliability, transcription accuracy, and multi-step reasoning in production customer service workflows.

A New Way to Make AI Actually Work in the Real World - May 14, 2026

IBM has launched Forward Deployed Units (FDUs), a new consulting delivery model that combines senior human teams with AI agents. This approach aims to accelerate enterprise AI deployment and ensure sustained value.

Research with immediate practical relevance

AI Produces First Comprehensive Models of Proteins in Motion

Researchers at École Polytechnique Fédérale de Lausanne (EPFL) have developed an AI framework that captures comprehensive all-atom protein dynamics, including side-chain flexibility. This advancement, detailed in a NeurIPS 2025 paper, moves drug discovery beyond static structure prediction.

AI models are getting better at replacing cybersecurity pros on certain tasks

The UK AI Security Institute (AISI) reports that the time horizon for frontier AI models to perform cybersecurity tasks is doubling every approximately four months, faster than previous estimates. Mythos Preview and GPT-5.5 models are significantly outperforming this trend.