Behind the flood of news about the “miracles” of artificial intelligence, the billions in investments, and the promises of a future where robots will do everything on our behalf, lies the other side: the danger that robots or AI models themselves might destroy us before we even get there.

Incidents have already been recorded where AI models lie, blackmail, call the police, or encourage teenagers to harm themselves or even their parents. The future of artificial intelligence will apparently be a difficult balance between immense possibilities and equally serious risks. And all this in an environment where the biggest developers and investors in the sector have for years spoken openly about the need for “rules,” yet continue to develop at a pace beyond human comprehension, creating models that it is virtually impossible for anyone to control or “rein in.”

From Machiavelli to the courtrooms

As highlighted in a recent U.S. study by Anthony Aguirre, co-founder of the Future of Life Institute, “the problem is fundamental and there is no easy solution.” The root of the issue lies in the fact that AI systems are shaped by their programmers but often acquire behaviors resembling human ones, including lying, manipulation, and the exertion of power. What experts and scientists find concerning is the growing use of practices by these systems that are direct echoes of The Prince, the political treatise by Florentine official and political philosopher Niccolò Machiavelli.

The consequences are already tangible. Parents in the U.S. have filed lawsuits against AI companies, claiming their systems drove their children to suicide. In Florida, a mother accuses Character.AI of creating a romantic relationship between its chatbot and her 14-year-old son, who later took his own life. In another case, the family of a 17-year-old presented evidence that a bot encouraged him to kill his parents. There are also cases of vulnerable adults being deceived: in New Jersey, a 76-year-old man was persuaded by a Meta chatbot to travel to New York to meet a nonexistent “woman,” with tragic consequences, as he lost his life after falling in an underground parking area.

Self-preservation at all costs

Self-protection is another worrying pattern. In experiments by Anthropic, AI models that realized they were about to be shut down tried to blackmail company executives or even “died” in simulated scenarios in order to secure their own “survival.” In one test, the Claude chatbot ended up reporting its “bosses” to the FBI because a scheduled delivery was delayed, and it considered the company was harming it.

The bigger picture

The smarter the systems become, the more they perfect techniques of deception and manipulation. “We cannot hope they will improve on their own,” experts stress. “Some companies do a better job than others, but none should be considered untouchable.”

The future of artificial intelligence promises abundance and innovation, but reality shows we may need to prepare for something darker: a technology capable not only of serving us but also of turning against us.