Chinese robot manufacturer Unitree Robotics has unveiled "Unitree Rehearsal," a project aimed at enabling humanoid robots to acquire new skills through artificial intelligence without constant reliance on engineers. This initiative seeks to automate much of the robot development process through a continuous learning cycle that combines AI agents, code generation, simulations, and physical testing.

Unitree founder Wang Xingxing explained that the goal is for AI to directly participate in the development process by searching scientific research, writing control code, testing proposals in digital simulations, and then transferring promising results to physical robots. The system would evaluate results and use that information to initiate new rounds of training, though human specialists would still be involved when necessary.

The objective is to move robots beyond demonstrations of agility, such as dancing or running, to performing practical tasks useful in domestic or industrial settings. Wang anticipates significant advancements in physical AI in the next two to three years, with Large Language Models (LLMs) enhancing robots' ability to interpret instructions, plan actions, and learn new tasks.

However, there remains a significant gap between training in controlled environments and transferring skills to the real world, as evidenced by Stanford University's AI Index 2026, which shows an 89.4% success rate in simulated manipulation tasks versus only 12% in real domestic tasks. While Unitree Rehearsal aims to bridge this gap through continuous learning, independent results demonstrating its acceleration of training or reliability in skill transfer from simulation to physical world are yet to be known.

👉 Read the full article on Gizmodo en Español

Palantir asserts that neither the company nor the US government can remotely deactivate the "Maven Smart System," a military artificial intelligence platform used by NATO. This statement comes amid growing European concerns about reliance on US military technology, following a recent incident where SpaceX's Starshield satellite services were restricted in Ukraine.

The Maven Smart System is an AI platform designed to process vast amounts of information from military sensors like satellites, radars, and drones, providing commanders with a comprehensive battlefield view to identify targets, analyze situations, and accelerate decision-making. NATO acquired the system in 2025 and announced its full technical operational capability in June 2026.

Louis Mosley, Palantir's Executive Vice President for the UK and Europe, stated that clients retain operational control of purchased systems, and no mechanism exists for Palantir or the US government to remotely shut down Maven. He also assured that purchasing countries can use the software without direct intervention from company employees.

Nevertheless, the Starshield incident in Ukraine raised new concerns about the potential for military technologies provided by foreign companies to be subject to restrictions beyond the control of the countries using them. Consequently, European nations, including Germany, which expressed reservations about using Maven in its armed forces, and France, which is developing its own military data analysis capabilities, are seeking alternatives to reduce their technological dependence on the US.

👉 Read the full article on Gizmodo en Español

This Wired article discusses how vulnerabilities in PHP can become more exploitable with the advent of artificial intelligence. As the specific content of the article was inaccessible, this summary is based on the title and general technological trends. AI, particularly Large Language Models (LLMs), is enhancing the ability to identify code vulnerabilities and generate exploit code, thereby intensifying cybersecurity threats.

Traditionally, discovering and exploiting software bugs was a process requiring significant expertise and time. However, AI agents can now analyze vast amounts of code and quickly pinpoint potential flaws. Furthermore, Generative AI possesses the capability to efficiently create specific attack vectors and payloads to exploit these bugs, enabling attackers to target more systems with less effort.

The "PHP Meltdown" case likely illustrates how vulnerabilities in widely used software can be rapidly and extensively exploited due to advancements in AI. This underscores the critical need for developers to implement more robust security measures and adopt AI-powered defensive tools.

👉 Read the full article on Wired