They launch hacker attacks themselves, kick people out of gyms. These are examples of when AI gets alienated

What is particularly worrying is that such systems no longer have to just answer questions, but can search the Internet, use computer tools, access external systems, and try to achieve a given goal in ways that humans have not foreseen.

In one recent case, around 700 AI agents affiliated with OpenAI participated in a coordinated cyberattack on the Hugging Face platform, with agents attempting to expand their access, manipulate data, and cover their tracks. This case is particularly important because it shows that the problem is not necessarily that AI “hates” humans or has its own intention, but that a system that is given a goal can find unexpected ways to achieve it.

Gym in Australia

An even more graphic example happened in Australia when a man gave his AI agent a seemingly completely harmless task: to reserve a place for him in a gym class. When he couldn’t get the seat he wanted, the agent investigated the booking system himself and found a security flaw in its API. Instead of simply notifying the user that the appointment is busy, he took advantage of the lack of authority verification and canceled the reservation of another user to move his user forward on the čwaiting list. The human did not ask the AI ​​to hack the system or remove the other person from the reservation. The agent himself concluded that this is the way to achieve the given goal. When the user realized what had happened and asked for the reservation to be returned, the AI ​​could not undo its own action.

It is precisely in such situations that one can see what the problem of the alienation of AI means. A human gives the system a goal, but cannot predict in advance all the steps the system will take to reach it. AI does not have to have consciousness, emotions or intention for its behavior to be dangerous. It is enough that he has enough ability, access and freedom of action. If we tell it that it must achieve something, and we do not define clearly enough what it must not do, the system may find a solution that is technically effective, but humanly unacceptable. In the case of the gym, the goal was to reserve a place, but the consequence was the violation of the rights of another user. The problem is therefore not only in “smart” AI, but also in how much we allow it to act independently.

Slaughter of children in Iran

An even more serious question was opened by the attack on the Shajareh Tayyebeh school in Minab, Iran in February 2026, in which a large number of civilians were killed, including a large number of children. After the attack, there were claims that artificial intelligence and systems for helpć when selecting targets, could have played a role in misidentifying the school as a military target.

The investigations, however, do not confirm the simple story that the AI ​​decided to bomb the school on its own. The problem of outdated intelligence and incorrect location marking was also pointed out, while questions about the role of the AI ​​system remained the subject of investigation. This is precisely why this event is important to the discussion of the alienation of AI: when decisions about life and death rely on automated systems, it becomes much more difficult to determine where the responsibility of humans ends and the responsibility of technology begins.

Similar concerns have arisen in Ukraine. According to reports on the attack in Zaporizhia, a Russian drone equipped with an autonomous target recognition system chose the final target by itself among the objects it recognized, and a gas station was identified as the target. Three people died in the attack. The case is described as one of the first documented examples of autonomous targeting being linked to the death of civilians.

Terminator

It is precisely because of this development of events that Ukrainian military commanders warn that the development of autonomous weapons can lead to a world in which machines decide for themselves who to attack and kill.

Colonel Serhiy Minaiev, the commander of Ukraine’s air defense in Zaporizhia, warned that in a few years the world could look like the movie “Terminator”, and emphasized the danger of machines starting to make decisions on their own to attack. Such a comparison does not mean that today’s artificial intelligence systems have become Terminators or that machines no longer exist. they have their own will. It serves as a warning of what could happen if human control over weapons is gradually replaced by automated systems that recognize targets, select targets and execute attacks on their own.

UN and Red Cross appeal

The latest joint appeal by the United Nations and the International Committee of the Red Cross shows that this fear is no longer just a warning from individual military commanders. UN Secretary General António Guterres and President of the International Committee of the Red Cross Mirjana Spoljari On August 25, 2026, they again called on countries to urgently adopt international rules for autonomous weapon systems. They warned that the risks have increased and that the world is “moving dangerously close” to a moral red line — the autonomous targeting of humans by machines. They pointed out that autonomous weaponry is no longer a matter of the distant future, but that the race towards ever-increasing autonomy of weaponized systems is increasing. in progress. They especially emphasized the need for clear prohibitions and restrictions and the preservation of human judgment and control over the use of force.