By Emilio Crespo
The business driven by Artificial Intelligence, incomparable to any other advance in industrial civilization, creates models that are increasingly more powerful and more difficult to control.
The very dangerous cyber pirate that hides Mythos from the company Anthropic, presented in May, is now joined by another apparently much superior AI, GPT-6 Astra. According to its creator, OpenAI, it is the beginning of the
discussed “general artificial intelligence”, that capable of acting and training itself. Scientists specialized in these technologies warn that they are on the verge of emulating reasoning
autonomous. And in a language, the code of their neural programming networks, which humans could not decipher. GPT-6 Astra represents a milestone in this matter.
The trumpeted superpowers of each new AI that North American laboratories launch on the market may be marketing strategies to boost their benefits, as many think. But the statements and manifestos, repeated in August, by experts and employees of these firms suggest that the threats are real.
In addition, there have already been several cases of AI models, including a Chinese one, that escaped the control of their developers and committed illegalities. In the last known incident, about 700
OpenAI agents, much less capable than Astra, coordinated on their own to infiltrate a German website. What they could not do was hide the records of their actions. When they use their own neural language, these excesses can be undetectable.
Meanwhile, in the US, AI companies and the Government do not agree on creating a framework of legality and security so that advanced Artificial Intelligence, or those who use it, do not pose a danger to society.