Artificial intelligence is rapidly penetrating medicine, education, industry, finance, government, and everyday life. It can process data, create texts and images, write code, and assist in solving complex scientific problems in a matter of seconds, as reported by [Entrevue](https://entrevue.fr/ru/intelligence-artificielle/pourquoi-certains-experts-redoutent-ils-que-lintelligence-artificielle-puisse-menacer-lhumanite/). At the same time, the development of technology raises concerns among some researchers. They are worried not only about existing issues—deepfakes, cyberattacks, fraud, or job displacement—but also about the potential emergence of much more autonomous systems whose capabilities may match or exceed human abilities in many areas. ## A Machine Doesn’t Have to Hate Humans One of the most discussed scenarios is related to the so-called goal alignment problem. Developers may set a task for the system, but the AI may understand or execute it differently than intended by humans. For a machine to pose a danger, it does not need to possess consciousness, emotions, or a hostile attitude towards humans. It is sufficient for a powerful system to consistently pursue a poorly defined goal while having access to critical resources or infrastructure. For example, a requirement to maximize a certain metric could lead to undesirable consequences if ethical, legal, and social constraints are not specified in the task. The more autonomous the system and the broader its powers, the harder it is to foresee all possible ways to execute the command in advance. ## The Main Fear—Loss of Control Some experts fear that the capabilities of AI will grow faster than methods of verification and regulation. Developers may encounter a system whose decisions cannot be fully understood, predicted, or adjusted in time. Particular concern arises from the hypothetical ability of advanced AI to participate in its own improvement. In the most radical scenario, each new version of the system would create an even more efficient version, leading to a rapid acceleration of technological development. It is still unknown whether such a development is possible in practice. However, proponents of enhanced safety measures argue that it is necessary to prepare for potential problems before extremely powerful systems emerge, as it may be impossible to rectify the situation after losing control. ## Real Threats Already Exist Significant harm to humanity does not require waiting for the emergence of superintelligence. Modern models are already capable of helping create convincing fake images, videos, and audio recordings, as well as mass-producing manipulative messages. Such tools can be used for fraud, political propaganda, election interference, and undermining trust in the media. When it becomes difficult to distinguish genuine recordings from artificially created ones, people risk losing trust even in real evidence. In the field of cybersecurity, AI can accelerate the search for vulnerabilities, prepare phishing emails, and conduct attacks. These same technologies assist specialists in protecting information systems, so the outcome largely depends on who uses them and for what purposes. ## Weapons and Decisions Without Human Involvement Specific concerns are associated with the use of artificial intelligence in the military sphere. Automated systems are already used for reconnaissance, image analysis, route selection, and identifying potential targets. The main ethical question is whether it is possible to trust an algorithm with decisions that affect human lives. A recognition error, incorrect data, or misinterpretation of the situation can lead to the deaths of civilians and unintended escalation of conflict. Experts therefore insist that the final decision to use lethal force must remain with humans. ## Another Risk—Concentration of Power AI can pose a threat not only due to its own errors or the actions of criminals. Powerful systems may end up in the hands of a limited number of companies, states, or individuals. Control over technologies allows influencing the information space, monitoring citizens, managing access to knowledge, and making decisions that affect millions of people. In the absence of transparency, society may not even know why an algorithm denied a person a loan, job, insurance, or social assistance. That is why experts speak of the need for independent audits, protection of personal data, and the ability to appeal decisions made with the involvement of algorithms. ## Can AI Destroy Humanity It is impossible to determine the likelihood of the complete extinction of humanity due to artificial intelligence. Systems that possess the necessary level of autonomy and control over the real world for such a scenario do not yet exist. Some researchers consider discussions about extinction too speculative and urge a focus on already observable problems—discrimination, surveillance, disinformation, and economic inequality. Others argue that even a very unlikely threat cannot be ignored if its potential consequences are global and irreversible. In their opinion, security research should evolve simultaneously with the creation of more powerful models. ## How to Mitigate Risks Experts suggest testing new systems before their widespread implementation, limiting AI access to critical infrastructure, and maintaining the possibility of human intervention. Equally important are monitoring the use of personal data and holding companies accountable for the consequences of their products. The creation of international rules is also being discussed, especially for military applications of artificial intelligence and the most powerful universal models. However, developing uniform standards is challenging: states strive to maintain technological advantage and are not always willing to disclose information about their own projects. The main danger of artificial intelligence lies not in the image of the "evil machine" from science fiction films. Real risks arise from poorly defined goals, excessive autonomy, algorithmic opacity, and malicious use of technology by humans. The scenario of losing control remains a hypothesis, but fraud, deepfakes, surveillance, and weapon automation require attention today. Therefore, the development of AI must be accompanied by safety checks, clear rules, and the preservation of human responsibility for critical decisions.
