Integral AI, a company focused on developing artificial intelligence that can move through the real world in the form of robots, has announced successful tests of a model called AGI Sandbox. According to the company, this is the world's first model to achieve the level of artificial general intelligence, abbreviated as AGI. This means that the system can learn new skills on its own without requiring pre-prepared datasets or human intervention. In tests, robots equipped with this model learned new tasks in real-world environments entirely on their own, without any supervision.
The company defined AGI using three main criteria that it must meet. The first is autonomous skill acquisition, in which the system independently discovers and masters new tasks in unfamiliar domains. The second criterion concerns safety and reliability—learning takes place without risks or unexpected problems. The third is energy efficiency, meaning that energy consumption during learning is equal to or lower than that of a human learning the same skill. Integral AI engineers used these criteria as key benchmarks during development and testing.
Who Is Behind Integral AI
The Integral AI team is led by Jad Tarifi and Nima Asgharbeygi, who previously worked on generative intelligence at Google. After leaving Google, they moved to Japan and now work outside the traditional Silicon Valley hub. Jad Tarifi, the CEO and co-founder, described the announcement as a technical achievement that opens a new chapter in human civilization. According to him, the goal is to scale this model toward superintelligence that would help expand freedom and collective capabilities.
The company specializes in embodied AGI, which is intelligence embedded in robots or machines that interact with the physical world. Its approach differs from conventional AI systems, which are often limited to specific tasks and depend on enormous amounts of data. Instead, AGI Sandbox mimics the multilayered structure of the neocortex, the part of the brain responsible for human thought. This architecture allows the system to grow, abstract information, plan, and act as a unified whole.
Development and Testing Details
During testing, AGI Sandbox proved highly adaptable, particularly in the field of autonomous robotics. Engineers observed robots equipped with the model independently acquiring new skills in real-world environments without any human intervention. This is a major difference from current AI models, which require large amounts of data and human assistance to learn.
The announcement came on December 8, 2025, and attracted interest because AGI has long been an anticipated goal in the field. Many experts are asking whether this definition of AGI will be widely accepted. Integral AI defined AGI itself to fit its model, which is a clever move, but time will tell whether it convinces others. In any case, it opens up new possibilities in robotics, where machines could learn on their own, without risks and with low energy consumption.
The AGI Sandbox model is still at an early stage, but the company plans to scale it toward fully embodied superintelligence. Jad Tarifi emphasized that this is the foundation for systems that will be scalable and self-improving. Compared with other companies such as OpenAI or DeepMind, which are also pursuing AGI, Integral AI focuses on safety and efficiency, addressing concerns about AI's environmental impact.
Tests confirmed that the system develops dynamically from within, enabling rapid adaptation. In robotic applications, it demonstrated the ability to solve unfamiliar problems in ways similar to humans, including playing games or adapting to new situations. This approach could transform how AI systems are developed by reducing dependence on human intervention and increasing autonomy.
Sources: techedgeai.com and aithority.com
