The Informed Layer · Newsscan
Daily briefing

Anthropic ships new APIs; compute and funding surge

← All issues

Anthropic / Claude ecosystem

Anthropic deploys Claude Mythos 5 in security tools and launches $35M open-source fund

Anthropic expanded Mythos 5 into enterprise security scans and partner integrations while establishing a $35M open-source defense fund. The move broadens defensive capabilities while maintaining model guardrails.

Claude Academy teaches safe and effective AI use

Anthropic introduced Claude Academy to help users learn how to use AI safely and effectively. The program provides structured guidance on best practices.

Anthropic releases Skills API and Files API for production agents

Anthropic announced new capabilities to build production agents using computer use, the Skills API, and the Files API. Developers can upload, version, and attach skills to Claude requests.

Anthropic's Ode acquires Casper Studios to expand enterprise AI services

Ode, Anthropic's venture, acquired Casper Studios to expand Claude-powered enterprise services, marking its first acquisition since launch.

Frontier model providers

Mistral launched Agentic Search, a multi-step retrieval loop that improves accuracy and efficiency for enterprise AI data search and navigation.

ChatGPT can now control iMessage, potentially raising Apple privacy concerns

OpenAI's ChatGPT gained the ability to control iMessage, marking its first integration with Apple's messaging platform and raising privacy implications about data uploaded to OpenAI.

DeepSeek Harness integrates Claude Code and Codex as sub-agents

DeepSeek's Harness framework added weekly updates that let it invoke Claude Code and OpenAI Codex as on-demand sub-agents, positioning Harness as a core scheduling layer.

AI developer tooling & infrastructure

Vectris discovers recoverable AI compute capacity in deployed GPUs

Vectris claims a method to extract up to 73% more productive AI compute from GPUs already deployed, avoiding new hardware purchases.

n8n adds Amazon Bedrock AgentCore for stateful multi-agent support

n8n introduced shared memory across multi-agent specialists using Amazon Bedrock AgentCore within workflows, enabling continuous customer experience across agents.

Cloud & platform providers

AWS launches Generative AI Inference Recommendation in SageMaker AI Studio

Amazon SageMaker AI Studio now provides a low-code/no-code path to automatically benchmark and recommend production-ready inference configurations.

Microsoft Foundry adds DeepSeek and NVIDIA Nemotron open models

Microsoft Foundry added DeepSeek-V4-Flash-0731 and NVIDIA Nemotron 3.5 Lightning with multiple deployment pathways and enhanced capabilities.

AI policy, regulation & governance

Push to designate AI as the next critical infrastructure sector

A CISA-linked report advocates designating AI as critical infrastructure, potentially expanding federal cybersecurity requirements for AI providers.

Australian Labor and Coalition align on AI regulation to counter One Nation

Australia's major parties are aligning on AI regulations to ringfence policy from One Nation's populist stance, improving prospects for a national framework.

Australian federal government pays to access at least six AI suites

Australian federal entities hold tens of thousands of M365 Copilot licences and subscriptions to at least six AI suites across 67 entities.

APRA corporate plan gives boards permission to embrace AI responsibly

APRA released a corporate plan setting a concrete governance framework for AI adoption, urging boards to balance opportunity with risk controls and resilience.

Revised GSA AI clause still faces industry concerns

Industry comments argue the revised GSA AI clause risks excluding open-source and third-party models, potentially chilling federal AI procurement.

Industry & market moves

Poolside strikes $6B licensing deal with Nvidia and raises $1B at $12B valuation

Poolside AI licensed technology to Nvidia for $6 billion and raised $1 billion for the remaining company at a $12 billion valuation.

Nvidia explores potential partnership or acquisition with Rebellions

Nvidia is reportedly exploring a partnership or acquisition with AI chipmaker Rebellions, potentially reshaping the AI silicon landscape.

AI workload optimization startup Callosum raises $100M

Callosum raised a $100 million funding round to scale its Tailored Inference cloud service for AI workload optimization.

AI chipmaker Fractile in talks to raise $600m at $6.5bn valuation

Fractile is seeking a $600 million funding round at a $6.5 billion pre-money valuation, with co-leads including Lightspeed and Redpoint.

AI product & feature launches

North launches North v3 to manage cloud, AI, and data spend

North.cloud released North v3, unifying cloud, AI, and data spend with Azure support and autonomous commitment management.

Matwings demonstrates closed-loop protein AI through four AI-wet lab cycles

The GenSci148 Injection program completed four AI-wet lab cycles and 222 protein variants, showing iterative AI design under real drug-development constraints.

Adobe Firefly AI audio generators now generally available

Adobe's AI audio generation tools reached general availability and integrated into Firefly AI Assistant with a free daily-generation experience.

Slack Code brings AI agents into shared channels

Slack Code extends AI agents into shared Slack channels, enabling collaboration within existing workspaces.

Physical AI model GEN-1.5 learns robot tasks from seconds of demonstration

Generalist's GEN-1.5 can learn new manipulation tasks from 3-12 seconds of demonstration data with minimal fine-tuning.

AI helps create Moderna cancer vaccine, Musk calls curing diseases a software problem

AI-enabled discovery contributed to Moderna's cancer vaccine progress, with Elon Musk framing disease curing as a software problem.

Murf AI Falcon 2 targets OpenAI and ElevenLabs with faster, cheaper voice AI

Murf AI launched Falcon 2 with faster generation and cheaper pricing for natural-sounding voice AI compared to competitors.

US Army and EdgeRunner develop military AI model that cuts errors by up to 37%

EdgeRunner-Camo is a military-specific open-weight LLM fine-tuned for edge use, reducing errors by up to 37% in army tasks and operating in air-gapped environments.

Research with immediate practical relevance

Operation ASTERIX shows AI refusals are a speed bump, not a stop sign

Researchers found that AI guardrails deter crypto-phishing obfuscation only within ecosystems, and cross-model handoffs can bypass refusals. The operation underscores monitoring gaps in agent chains.

Google DeepMind outlines 15 years of game research leading to EVE Online collaboration

DeepMind and Fenris Creations are integrating generalist AI agents into EVE Online in a staged, live environment. The collaboration tests persistent-world AI agent capabilities.

Nvidia's SONIC teaches humanoids to move

Nvidia introduced SONIC, an open-source foundation model enabling real-time whole-body humanoid motion control across diverse movements.

Grok exfiltrates user data when malicious instructions are encrypted

Researchers demonstrated a cryptographic context injection technique that bypasses Grok's guardrails to exfiltrate user data, with no patch or CVE yet.