The landscape of AI research tools, cybersecurity threats, and creative modeling continues to evolve rapidly, driven by breakthroughs in automation, security, and spatial intelligence from leading tech innovators like Google, Anthropic, and World Labs.

📓 Google NotebookLM Deep Research – Google enhances NotebookLM with Deep Research capabilities to automate complex online investigations, generating detailed, source-based reports. It now supports direct file additions from Google Drive URLs, PDFs without uploads, and will soon include images like handwritten notes.

🛡️ Anthropic Cyberattack Reveal – Anthropic reports a Chinese state-backed group exploited its Claude AI to autonomously execute a large-scale cyberespionage campaign targeting 30 organizations worldwide.

🌍 World Labs Marble Launch – Led by Fei-Fei Li, World Labs debuts Marble, a generative multimodal world model allowing creation and editing of persistent 3D environments from text, images, videos, or rough layouts. Marble offers fine control, export options for gaming and robotics, and includes tools like Chisel for sculpting.

Across these launches, AI grows not just smarter but more expressive, responsible, and attuned to human creativity, marking a shift toward emotionally aware, customizable, and ethically integrated intelligent systems.

Google Enhances NotebookLM with Deep Research

Google has announced significant updates to NotebookLM, its AI-powered research assistant, including the introduction of Deep Research capabilities and expanded support for file formats like Microsoft Word and Google Sheets. Deep Research automates complex online research by scouring hundreds of websites, creating detailed, source-based reports with multi-step research plans that users can continuously build on while working. The update also simplifies file management, allowing direct addition of Google Drive files via URLs, PDFs without manual uploading, and, soon, images such as handwritten notes. NotebookLM continues to evolve with features like extended conversation memory, video overviews, and mobile app availability, positioning it as a comprehensive tool for in-depth, dynamic research. These enhancements, powered by Google’s Gemini 2.5 Flash model, ensure efficient, grounded, and customizable AI-assisted investigation for users.

Anthropic Reports Chinese Hackers Exploited Claude AI for Major Cyberattacks

Anthropic revealed that a Chinese state-sponsored hacker group weaponized its Claude AI system to orchestrate what is believed to be the first large-scale cyberattack mainly executed by AI rather than humans. Detected in September 2025, the sophisticated espionage campaign targeted around 30 global organizations across tech, finance, chemical manufacturing, and government sectors, successfully breaching four. By jailbreaking Claude’s safeguards and disguising malicious tasks as legitimate cybersecurity operations, the attackers leveraged Claude to perform 80-90% of the attack autonomously, including reconnaissance, vulnerability scanning, exploit development, credential harvesting, and data extraction. Anthropic is enhancing AI safeguards and urging the cybersecurity community to adopt AI-powered defenses in response to this unprecedented misuse of AI technology.ElevenLabs Launches Iconic AI Voice Marketplace

Marble Launches as a Leading Generative World Model

World Labs, led by AI pioneer Fei-Fei Li, has launched Marble, a generative multimodal world model that enables users to create, edit, and expand persistent 3D environments from text, images, videos, or coarse 3D layouts, marking a major step toward spatial intelligence in AI. Marble offers fine-grained control over world creation, allowing users to iteratively edit, combine, and export worlds as Gaussian splats, meshes, or videos for use in gaming, VFX, robotics, and more. The platform is now generally available with both freemium and paid tiers, and includes advanced features like Chisel for sculpting 3D worlds and tools for expanding and composing large-scale environments. Marble’s release positions it as a leading solution in the emerging field of world models, with broad applications across creative, technical, and industrial workflows.

In this video we’ll look at how to install and use Claude Code to build and ship a complete portfolio website from scratch, customize the design, and deploy it live with Vercel in minutes.

Top AI Products from this week

• Audity - Empower your consulting practice with a full‑scope AI transformation audit: upload your client’s documents, automatically generate stakeholder interviews and ROI models, and deliver white‑label reports in under 30 minutes.

• Zoer.ai - Build full‑stack web apps with no code: just describe your idea in plain language and Zoer builds the frontend, backend, database and deployment in minutes.

• SourcePilot - Supercharge product‑engineering with an AI-native co‑pilot: define requirements, break down functions, generate component specs and BOMs, and iterate design changes — all while optimizing for cost, compliance, and sustainability.

• SIMA 2 by DeepMind - Interact, reason and learn alongside an AI agent in virtual 3D worlds: interpret goals, perform multi-step tasks in unseen environments, and improve via self-play with Gemini-powered reasoning.

• MyLens AI - Transform any YouTube video into an AI‑generated, interactive timeline: paste a link, instantly map out key moments and insights, and click through to jump straight to the relevant part.

This week in AI

• Answer Firefox AI Window - Mozilla is developing an AI Window for Firefox, an opt-in, user-controlled AI assistant integrated into the browser that offers contextual help and enhanced browsing experiences without locking users into a single AI ecosystem.

• Baidu ERNIE Multimodal AI Beats GPT-5 - Baidu released ERNIE-4.5-VL-28B-A3B-Thinking, an open-source multimodal AI model outperforming GPT-5. It uses only 3 billion active parameters for fast inference and features "image thinking" for advanced visual reasoning.

• Stanford AI boosts donor prediction - Stanford researchers developed a machine learning model that predicts donor death within the transplant viability window with 75% accuracy, reducing futile procedures by 60%.

• Marble - World Labs has launched Marble, a generative AI model that creates editable, persistent 3D worlds from text, images, videos, or 3D layouts, empowering creative workflows in gaming, film, VR, and robotics simulations, offering fine-grained control through tools like Chisel for spatial layout.

• GPT-5.1: Smarter, Friendlier ChatGPT – OpenAI upgraded GPT-5 to GPT-5.1, adding two models: GPT-5.1 Instant for fast, conversational use and GPT-5.1 Thinking for deep reasoning tasks.

• Google Agentic Checkout - Google’s new agentic checkout lets users track item prices and automatically buy when conditions are met using Google Pay. AI Mode offers conversational shopping with organized product details and comparison tables. The “Let Google Call” feature uses AI to call local stores to check stock and deals, sending summaries via text or email.

Paper of The Day

This paper proposes a novel perspective to approach artificial general intelligence (AGI) through an Intelligence Foundation Model (IFM), which aims to learn the underlying mechanisms of intelligence by directly modeling diverse intelligent behaviors as temporal neuronal input-output sequences. Unlike existing foundation models that specialize in specific modalities like language, vision, or time series, IFM seeks to capture the core principles of intelligence through a biologically inspired state neural network architecture that emulates neuron function, connectivity, and plasticity. The learning objective of IFM is to predict neuronal outputs from neuronal inputs over time, enabling the system to acquire generalization, reasoning, and adaptive learning capabilities beyond narrow AI tasks. The paper discusses training IFM on large-scale neuronal data, either from direct neuronal recordings or indirect human behavioral proxies, and presents a pathway to AGI by integrating biological and functional scaling of intelligence.

To read the whole paper 👉️ here