March 2026

Anthropic, the safety-first AI company behind Claude, accidentally exposed details of its next big model, "Claude Mythos," through a website error; unpublished blog posts and specs ended up public. They call it a "Capybara" tier, bigger and smarter than their top Opus model, with huge jumps in reasoning, coding, and especially cybersecurity skills. From a […]

Agent-driven work means AI agents act as full operators, not just helpers—they handle complex tasks like decision-making, data processing, and operations 24/7, like a team of digital employees. OpenClaw's launch a few months ago sparked this shift by proving agents could tackle real-world business challenges autonomously. We reached this stage through rapid AI evolution: from

A group of Iran‑linked hackers, calling itself the “Handala Hack Team”, has publicly claimed to have breached FBI Director Kash Patel’s personal email account and published photos, early‑career documents, and old emails online. The attack appears to be a direct retaliation for the U.S. government seizing the hackers’ online domains and offering a large reward

Meta has unveiled TRIBE v2, an open-source AI model that predicts how your brain neurons fire when you watch videos, hear sounds, or read text. Trained on over 1,000 hours of fMRI brain scans from more than 700 people, it simulates activity across 70,000 brain regions — far surpassing the noisy, limited real scans hampered

ARC-AGI-3 is a fresh benchmark from François Chollet's ARC Prize Foundation, designed like fun puzzles or games. Humans glance at colorful grids and figure out hidden patterns to solve them — scoring 100% on the first try. But AI must learn the rules from scratch, without hints, testing raw smarts like discovering goals and planning

OpenAI’s decision to wind down its Sora video generator shows a sharp shift away from “side quests” toward its next flagship model, Spud. Sora, which briefly topped the App Store and had a headline Disney deal, is being shut so its compute can be redirected. The Sora team will now work on “world simulation” for

Remote agents like OpenClaw have ignited the AI world by pioneering open-source autonomy, enabling LLMs such as Claude or GPT to control computers, manage workflows via messaging apps like Telegram, and execute real tasks locally. It was created by Peter Steinberger, who joined OpenAI after its viral explosion to over 100,000 GitHub stars. OpenClaw shattered

China is supercharging its robotics scene with OpenClaw, the open-source AI agent that's already dominating laptops worldwide. Developers are wiring it into Unitree's agile G1 humanoids, enabling real-time navigation and natural language commands—like "pick up that box" executed flawlessly in physical spaces. Ecovacs, a robotics powerhouse, just unveiled Bajie, a sleek home bot powered by

Robotaxis have transformed urban mobility in San Francisco, where services like Waymo and Cruise now operate 24/7, delivering smooth, rule-following rides that excite users with their precision and affordability compared to traditional options. Waymo handles complex scenarios effectively, though incidents like Cruise's past mishaps highlight ongoing safety tweaks. Further expansion still faces steep hurdles. Key

Anthropic's groundbreaking study, using Claude to conduct 80,508 open-ended interviews across 159 countries and 70 languages in just one week, reveals a nuanced public view of AI beyond simplistic polls. Top hopes center on professional excellence (19%), where AI handles routine tasks for strategic focus, alongside time freedom, life management, and personal growth. Fears like

Amazon’s new phone project, codenamed “Transformer”, is less about beating the iPhone and more about turning the smartphone into an AI‑driven Alexa hub. Led by former Microsoft‑Xbox‑Zune executive J Allard under Amazon’s ZeroOne unit, the device is being built around Alexa and Amazon’s broader ecosystem, from shopping and Prime Video to smart‑home and car‑linked services.