AI News

AI no longer responds — it acts. OpenAI launched GPT-5.6 with parallel agents and bidirectional voice, xAI released Grok 4.5, and Meta Muse Spark 1.1. The chatbot is turning into infrastructure — and whoever controls the loop between intent and outcome will own the next decade.

While Anthropic's Claude model sets the standard for logic and safety, OpenAI and xAI are charting their own paths. Let's take a look at how these approaches differ and what it means for users in Europe.

What do a chess prodigy from London, defeating the world champion in Go, and the Nobel Prize in chemistry have in common? The Infinity Machine by Sebastian Mallaby reveals for the first time the complete story of Demis Hassabis — a man who doesn't found companies for money, but out of an obsession with understanding intelligence itself.

SpaceXAI has introduced the Grok 4.5 model, which moves beyond mere conversation to autonomous task execution. It focuses on complex programming and the ability to act as a digital agent.

AI chatbots provide 73% higher quality health information than Google. A study from July 11, 2026 compared ChatGPT, OpenEvidence, and traditional search in answering questions about 13 breast pathologies. ChatGPT scored 62.2 out of 80 points, while Google managed just 35.9.

CES 2026 in Las Vegas showed that humanoid robots are no longer just laboratory experiments. Agibot brought three new product lines, Unitree had its machines box, and Sharp demonstrated a hand with 22 degrees of freedom.

Boris Cherny, creator of Claude Code, stated that not a single line of human-written code remains anywhere at Anthropic. The company is experiencing an internal upheaval — but at the same time, it's facing a crisis of trust that's driving developers in droves to OpenAI's competing Codex.

Meta's new Muse model pushes the boundaries of image generation, but it also raises a burning question: where does it draw from? Find out how your public Instagram photos help train AI — and what you can do to protect yourself.

Rubrik launches the first security platform built on agentic AI. The new system promises to cut recovery time after a cyberattack from weeks to minutes thanks to autonomous agents that independently plan and execute entire recovery processes.

An international team of researchers has demonstrated the first multi-stage attack on a quantum neural network running on real hardware. The attackers only needed to eavesdrop on power consumption and exploit crosstalk between qubits. The study shows that even today's imperfect quantum computers are vulnerable.