The Informed Layer · Newsscan
Daily briefing

Meta opens Muse; AI rules tighten worldwide

← All issues

Anthropic / Claude ecosystem

Anthropic expands independent evaluation with employee-level METR access

Anthropic announced an update expanding independent safety evaluation, granting METR and other external evaluators employee-level access to models and internal processes while maintaining controls over sensitive information.

Anthropic prepares Claude for the 2026 US midterms

Anthropic announced election safeguards for the 2026 US midterms: Claude now surfaces official voting information from Democracy Works and displays up-to-date election results within conversations.

Claude Code desktop adds nested sessions and saved environments

Claude Code desktop introduced nested sessions and saved environments, letting developers manage multiple parallel tasks with a clearer session overview via a sidebar.

Claude Code 2.1.299 stops sending gateway API keys to Anthropic

Claude Code 2.1.299 fixed an issue where the gateway's API key was transmitted to Anthropic when Claude Code was pointed at a third-party gateway, closing a potential credential exposure path.

Anthropic reports $16.8B quarterly revenue ahead of planned IPO

Bloomberg reports Anthropic's quarterly revenue hit $16.8 billion ahead of a planned stock market debut, with the company flagging profitability and safety commitments to potential investors.

Frontier model providers

GPT-4o users get three months' notice despite OpenAI's six-month deprecation promise

OpenAI is migrating GPT-4o API users to GPT-6 Luna with only three months' notice, contrary to its announced six-month minimum deprecation window, with notable changes in reasoning and tool-calling behavior.

SentinelOne cuts malware analysis from weeks to hours with OpenAI Daybreak

SentinelOne is using OpenAI's Daybreak to reduce malware reverse-engineering time from weeks to hours, expanding its threat investigation capacity.

OpenAI plans Hindi-tuned voice model and India-first UX

OpenAI plans a Hindi-tuned voice model plus India-focused, Android-first and connectivity-aware product experiences to accelerate adoption in the Indian market.

Meta unveils Muse Spark Open, top US open-weights model that runs on two paired Mac desktops

Meta released Muse Spark Open, an open-weights agentic coding model positioned as the new top US open-weights model, with enterprise strengths and modest hardware requirements (two paired Mac desktops), while keeping its frontier Muse Wonder closed.

Apple and Meta end long feud to bring Muse to Macs

Apple and Meta have ended their long-running feud with a rare collaboration to bring Meta's Muse AI to Mac computers.

Google expands News AI pilot to over 400 publications including US local news

Google's News AI pilot expanded to more than 400 publications, including US local news outlets, enabling AI-powered features such as story previews and audio briefings.

ChatGPT rolls out 'Intelligent UI' with interactive charts, forms and buttons

OpenAI's ChatGPT update fills responses with interactive visuals, including diagrams, charts, forms and tappable buttons, changing how conversational output is presented.

AI developer tooling & infrastructure

Amp puts AI agents to work beyond code with OpenAI models

Amp announced an expansion of its AI agents beyond coding tasks, now powered by OpenAI models for broader work use cases.

IntelliJ IDEA's Java and Kotlin engine for Cursor and VS Code hits 1.0

JetBrains released version 1.0 of 'Java and Kotlin by IntelliJ IDEA', bringing IntelliJ's core language intelligence to VS Code and Cursor.

Precisely launches MCP Server connecting AI assistants to SAP automation

Precisely introduced a standards-based MCP Server for Precisely Automate, connecting AI assistants to SAP automation via OData and MCP while preserving governance and Clean Core alignment.

Anchor launches Action Index, cutting computer-use agent token costs by 100x

Anchor's Action Index converts proven web workflows into reusable deterministic code, dramatically reducing computer-use agent cost and time per task.

Graaph launches MCP for code knowledge graphs aimed at AI coding assistants

Graaph launched graaph.org, providing interactive code knowledge graphs, MCP access and guides for AI coding assistants in a single site.

Cursor Origin expands with 20+ integrations and public app marketplace

Cursor's Origin now supports over 20 integrations and a public app marketplace, extending AI development workflows beyond core model performance into ecosystem connectivity.

Cloud & platform providers

BigQuery ships GA generative AI cost-management quotas

Google Cloud announced general availability of daily quotas for BigQuery generative AI functions, with per-project and per-user controls and monitoring for token-based costs.

Cloudflare ships fully programmable secure web gateway built on Workers

Cloudflare introduced a programmable secure web gateway architecture that separates TLS termination from Worker-based transformation logic, enabling custom rules and actions without hosting a traditional HTTP proxy; Gateway and Workers integration is now in public beta.

Cloudflare and OpenAI collaborate to improve AI discovery of the web

Cloudflare and OpenAI announced a joint effort to make AI discovery better for content creators and the agentic web.

Cloudflare adds DLP match validation in Workers

A Cloudflare changelog entry adds the ability to validate data loss prevention matches within Cloudflare Workers.

AI policy, regulation & governance

FSB finalises sound practices for responsible AI adoption in finance

The Financial Stability Board finalised sound practices for the responsible adoption of AI, providing practical governance guidance for safe adoption and resilience in the financial system.

EU publishes implementing decision on harmonised standards for high-risk AI systems

The EU Official Journal published an Implementing Decision on harmonised standards for high-risk AI systems under the AI Act, clarifying compliance expectations for regulated AI systems.

NIST draft guide on AI system evaluations praised by civil society with asks for clarifications

Civil society groups say NIST has made solid progress in its draft guide on AI system evaluations under the risk management framework, while urging clarifications on scope, definitions and coverage.

Australia introduces Digital Duty of Care legislation to parliament

Australia's Digital Duty of Care bill was introduced to parliament, imposing a duty on platforms to mitigate harmful content and granting users control over algorithmic feeds, with stronger eSafety Commissioner powers.

Bill Gates gears up for massive AI safety push

Axios reports Bill Gates has shifted from AI optimism to urgency and is preparing a major AI safety push calling for robust regulation and new regulatory institutions.

OpenAI agrees not to use trade secrets it allegedly took from Apple

In a legal settlement concession, OpenAI agreed not to use trade secrets it allegedly took from Apple and to disclose forensic data, resolving a patent/trade secrets dispute.

Barrett bill proposes Cabinet-level AI threats office (H.R. 10768)

Representative Barrett's bill, H.R. 10768, would create a Cabinet-level White House office to centralize AI and emerging technology risk oversight and advise the president on AI policy.

Industry & market moves

CERN and Mistral partner on a Large Physics Model for collider science

CERN and Mistral announced a partnership to create a single Large Physics Model trained on LHC simulations for multiple physics tasks, the first large-scale European collaboration of its kind for AI-driven science.

New Material joins Mistral to build enterprise frontier AI solutions

New Material is joining Mistral in an acquihire-style move to deliver design-driven, enterprise-grade frontier AI solutions and establish a Solutions Design team for closer customer collaboration.

AISLE and Cohere to join forces on autonomous AI cybersecurity

AISLE announced a cross-company acquisition with Cohere to advance autonomous AI cybersecurity capabilities.

Koch's venture arm backs a16z-backed Emanate's AI push into industrial sales

Koch Disruptive Technologies invested in Emanate, the a16z-backed startup bringing AI to industrial sales, confirming the deal in a Fortune exclusive.

CNN announces AI partnership with Google

CNN and Google announced a partnership on generative AI features to expand access to CNN journalism while protecting CNN's intellectual property.

Cursor stopped bundling Claude as Anthropic courts a $2 trillion listing

Cursor has paused bundling Claude in its subscription, a pricing and partnership shift with implications for Anthropic's distribution ahead of its potential $2 trillion IPO listing.

Microsoft reorganizes Azure business unit around AI under Arun Ulag

Microsoft is consolidating Azure divisions and realigning the cloud business around AI, with Arun Ulag taking charge of Azure Core and related executive teams.

Synopsys and Amazon sign multi-year, >$1B strategic IP agreement for AWS custom chips

Synopsys and Amazon reached a multi-year strategic IP agreement, reportedly over $1 billion, to accelerate AWS custom chip development using silicon IP and AI design tools.

Nvidia plans investment in inference chip rival d-Matrix

Nvidia plans to invest in rival AI inference chipmaker d-Matrix, expanding its ecosystem play and improving cross-compatibility with other chip designers.

OpenAI in talks for UAE capital in $30B funding round with possible $10B consortium

OpenAI is reportedly in talks for UAE capital to participate in a $30 billion funding round, with multiple funds potentially forming a consortium to inject up to $10 billion and BlackRock possibly joining.

Nvidia pledges $250M of compute to UK AI safety and plans stakes in Boltz, Foresight and Hiverge

Nvidia pledged $250 million of compute to UK AI safety efforts and plans strategic stakes in Boltz, Foresight and Hiverge.

Nvidia backs Doctolib, Genesis AI, Pigment and Entalpic as Huang flags $2B more for Europe

Nvidia announced backing for Doctolib, Genesis AI, Pigment and Entalpic, with Jensen Huang flagging an additional $2 billion for European AI investment.

AI product & feature launches

Superwhisper and Cohere partner on local meeting recorder for enterprise

Superwhisper partnered with Cohere on a meeting recorder that runs AI models locally, enabling private transcription without sending data to the cloud.

Nvidia expands Nemotron family with open Super VL and Bolt models from data center to desktop

Nvidia released Nemotron 3.5 Super VL and Nemotron 3.5 Bolt models (74B-A9B 24GB and E4B 4GB), open and optimized models covering server to edge deployments for agent workloads with post-training on users' own infrastructure.

Institute of Foundation Models and Inception42 launch Jais 3 Arabic-native AI

The Institute of Foundation Models and Inception42 launched Jais 3, a fleet of six open Arabic-native models plus Jais Speech for Arabic speech recognition, optimized for dialects and everyday Arabic use across speech, writing and agents.

Workday opens live data querying to Claude Code and Cursor agents

Workday opened headless access via its Trusted Agent Platform, letting agents built in Claude Code and Cursor query live Workday data with governance, audit trails and user-permission binding.

Shopify opens its catalog and checkout to any AI agent via Shop MCP

Shopify enabled AI agents to search catalogs, build carts and complete purchases within the agent via Shop MCP, reducing drop-off and bypassing merchant sites for signed-in shoppers.

Amazon launches Bee personal AI wearables that listen and take notes

Amazon unveiled new 'personal AI' devices, Bee Frontier and Bee Eclipse, hands-free wearables that transcribe conversations and integrate with email and calendar, with a subscription tier for advanced features.

Research with immediate practical relevance

Gemini 4 Argon used to train humanoid robots from video

A Korean report describes using Gemini 4 Argon to build virtual training environments from observed human actions, dramatically reducing the time and cost of training humanoid robots.

DeepSeek apologises for ~5-hour outage, offers full 7-day API spend as credit

Following a rare outage that took down web, app and V4.1-Flash API services, DeepSeek apologised and offered affected users a refund of their API consumption over the past 7 days as bonus credit; service has been restored.

Cursor staff investigated whether confidential data reached SpaceX for AI training

Cursor staff investigated whether confidential customer data surfaced in SpaceX's AI training, raising data-use scrutiny and corporate accountability questions.

Cloudflare Workers AI suffers partial outage

Cloudflare's official status page confirmed a partial outage affecting Workers AI on October 15.

Meta runs health data science research sprint with Muse models

Meta AI described a four-week collaborative research sprint using Muse models to accelerate hypothesis generation, coding and manuscript preparation with de-identified real-world health data.

SwarmRSI matches benchmark SOTA on 44.4% of NatureBench tasks

A new swarm-based recursive self-improvement approach, SwarmRSI, matched benchmark state of the art on 44.4% of NatureBench tasks with strong OpenProblems and MLE-Bench Lite results in an Oct 14 preprint.

LabInstruct benchmark shows video models fail lab procedure tests

The LabInstruct benchmark reveals a large gap between visual realism and procedural correctness in AI-generated lab demonstrations, challenging their use in self-driving-lab workflows.

Israeli researchers disclose high-severity flaw in Claude Code Desktop, since patched

Israeli researchers uncovered and disclosed a high-severity security flaw in Claude Code Desktop, which Anthropic subsequently patched.

Torvalds lets Claude Code tackle the Linux page cache; reviewers find a race

Linus Torvalds allowed Claude Code to work on the Linux page cache, producing a 10-patch series with a 23x throughput improvement in a synthetic benchmark, though reviewers found a race condition and real-world gains may be smaller.

New Aoki infostealer targets AI coding assistant credentials

A new infostealer dubbed Aoki specifically targets AI coding assistant credentials, marking AI access tokens as a valuable attacker target alongside wallets and passwords.