The Informed Layer · Newsscan
Daily briefing

Anthropic hardens models; policy pressure builds

← All issues

Anthropic / Claude ecosystem

Anthropic details strengthened defenses against distillation attacks

Anthropic published research on strengthening model defenses against distillation attacks, aimed at protecting frontier models from capability extraction. The work is part of Anthropic's broader safety and security research stream.

Anthropic licenses and releases Opaque in acquihire for confidential AI

Anthropic has acquired Opaque in an acquihire, licensing its technology and releasing the team to bolster Anthropic's confidential AI, on-prem and sovereign AI capabilities. Constellation Research reported the deal.

Claude Code's Anatomy skill visualizes technical ideas as interactive diagrams

A new Anatomy skill for Claude Code generates interactive isometric diagrams to visualize technical concepts, aiming to improve comprehension over traditional flowcharts. It extends Claude Code's visualization tooling for developers.

Frontier model providers

Meta launches AI tool letting users shape their Facebook feed with prompts

Meta introduced a 'Your Algorithm' feature that lets users reshape their Facebook feed and Reels using natural-language prompts. It is a user-facing application of AI-driven feed control.

xAI introduces unified X Subscription bundling Grok, Cursor and X Premium

xAI launched the X Subscription, a unified plan combining Grok, Cursor and X Premium with tiered usage pools and expanded enterprise controls. It consolidates xAI's consumer and developer AI offerings under one pricing structure.

AI developer tooling & infrastructure

VS Code 1.141 unifies Copilot and Codex sessions, tightens agent security

VS Code 1.141 brings unified session management across Copilot and Codex agents, cross-app conversation continuity, and enhanced sandboxing and security for AI agents. The release targets safer multi-agent development workflows.

Cloud & platform providers

AWS brings EC2 G7 GPU instances to Frankfurt and Stockholm

Amazon EC2 G7 GPU-accelerated instances are now available in the Europe (Frankfurt) and Europe (Stockholm) regions, delivering up to 4.6x AI inference and 2.1x graphics performance versus G6. The expansion targets European AI workloads with data-residency needs.

AWS launches Lambda Web Functions for HTTP-native serverless web apps

AWS announced Lambda Web Functions, a new HTTP-native Lambda form factor handling multiple concurrent requests with active-CPU billing. It reduces complexity for web apps and APIs running on Lambda.

Cloudflare pipes production issues straight to coding agents

Cloudflare introduced a capability that routes production issues directly to coding agents, aiming to close the gap between detection and remediation. The feature reduces manual handoffs in incident response workflows.

AI policy, regulation & governance

Newsom announces results of California's first-in-the-nation AI public engagement project

Governor Newsom announced results from California's first-in-the-nation project involving Californians in decisions about AI. The initiative gathered public input to inform state AI policy.

Pre-vetting accusation roils Australian AI inquiry as politicians grapple with AI

The Canberra Times reports accusations of government pre-vetting of the independent AI inquiry's recommendations, raising concerns about political interference. The dispute adds friction to Australia's AI governance debate.

Albanese government seeks AI security boost as scientists call for global oversight

The Albanese government is considering a boost to AI security funding as scientists call for global oversight of frontier AI. The push adds a security-investment dimension to Australia's AI agenda.

Kupor: 'Super Intelligence' task force will press companies on AI safety

White House task force chief Kupor said the 'Super Intelligence' task force will pressure AI companies on safety practices, in a Politico interview. The statement signals a more assertive posture from the task force formed earlier this month.

EU AI Office seeks information from developers after loss-of-control incidents

The European AI Office has requested information from frontier model developers following loss-of-control incidents involving advanced models. The move marks a shift from theoretical risk assessment to evidence-gathering on concrete incidents.

Australia's AI Review Committee sets out expert advice on public sector AI use

Australia's AI Review Committee has set out expert advice on public sector AI deployment, emphasizing human oversight, privacy and scalability. The advice is non-binding but signals future regulatory direction.

Industry & market moves

Workday and Anthropic announce partnership connecting Claude reasoning to business rules

Workday and Anthropic announced a strategic partnership to integrate Claude reasoning with the business rules that run enterprise processes across HR and finance workflows. The tie-up embeds Claude deeper into enterprise system-of-record operations.

Google DeepMind plans robotics hardware lab near Boston

Google DeepMind is planning a dedicated robotics hardware lab near Boston in Massachusetts to test and develop embodied AI systems. The move signals DeepMind's shift from purely model research toward hardware-enabled robotics.

Mistral AI expands enterprise cloud partnerships across Europe

Mistral AI is broadening partnerships with European enterprise cloud providers to deploy its sovereign generative models across the continent. The expansion targets regulated European organizations seeking EU-hosted AI.

Rocket Software acquires Geniez AI to accelerate agentic mainframe modernization

Rocket Software announced the acquisition of Geniez AI to accelerate agentic AI for mainframe modernization, deploying enhanced EVA capabilities with governance and control. The deal targets legacy enterprise modernization with agents.

Workday to acquire C1.ai, an agentic identity platform for people and agents

Workday announced an agreement to acquire C1.ai, a leading agentic identity platform, positioning Workday as an identity governance layer for people, money and AI agents. The deal secures real-time access control across enterprise systems.

Agilent to acquire Proscia for AI-enabled digital pathology

Agilent announced an agreement to acquire Proscia, expanding its digital pathology platform with Proscia's AI-enabled tissue diagnostics software. The deal strengthens Agilent's position in AI-driven diagnostics and companion diagnostics.

Napster and HUMAIN partner to deliver Arabic-native AI across the region

Napster and Saudi AI firm HUMAIN announced a partnership to bring Arabic-native AI to public and private organizations across the region. The deal embeds Saudi-developed AI capabilities into an international platform.

73 Strings acquires Callisto to boost AI in private markets

73 Strings acquired Callisto, integrating its AI engineering team to accelerate private markets technology with expanded data automation and scenario analytics. The deal combines acquisition with product integration.

AI product & feature launches

Nature Will's Natura-1 lets robot hands learn dexterity from a million hours of human video

Robotics lab Nature Will unveiled Natura-1, a model trained on roughly a million hours of first-person human hand video that adapts dexterous skills to robot hands with limited robot practice. Demo videos show robot hands folding wontons.

IFS launches IFS Intelligence governed by an Industrial AI Harness

IFS launched IFS Intelligence, providing live-operational reasoning from IFS and third-party data through a single conversational interface, governed by a new Industrial AI Harness. The platform targets industrial and field-service operations.

Kakao evolves KakaoTalk into an AI assistant with Kanana Space

Kakao is expanding its AI capabilities with Kanana Space, which understands context and preferences to enable natural voice-driven tasks such as marking messages read and ordering lunch, alongside a broader IP-based character ecosystem.

Liquid AI releases open-weight LFM2.5-24B-A2B for on-device agents

Liquid AI released LFM2.5-24B-A2B as an open-weight model, the largest and final model in the LFM2.5 family with a doubled context window and expanded vocabulary. It targets on-device agentic workloads.

Fish Audio launches Drama 3 with plain-language voice control and mid-sentence emotion shifts

Fish Audio launched Drama 3, enabling direct AI voice generation in plain language with word-level control and emotion shifts mid-sentence. The release targets granular, real-time expressive voice synthesis.

Research with immediate practical relevance

CONTRA framework lifts selective clarification in coding agents, ships Claude Code plugin

Researchers released CONTRA, a training-free selective clarification framework for LLM code generation that improves question-validation F1 by 13.88%. It ships with an open-source Claude Code plugin and is documented in an arXiv paper.

Microsoft's Code Video Model plans video shots in code first

A Microsoft project introduces a three-step proxy-to-video pipeline in which shots are planned in code first, preserving structure while transferring appearance from a reference image without modifying the underlying MiniMax-H3 model.

Behind AI industry's call for independent reviews, a fight over what gets made public

Reporting reveals ongoing negotiations between AI labs including Anthropic and OpenAI and independent evaluators over whether and how evaluator findings can be published while labs protect trade secrets. The outcome will shape oversight of powerful AI models.

Princeton study: AI is great at math but lacks research judgment

Princeton Engineering researchers present a case study in which a two-model AI system narrowed Grothendieck constant bounds with human guidance, concluding that AI can solve math problems but cannot yet build mathematical theories autonomously. The finding delineates current AI research capability limits.