Bill Gates, one of the founders of Microsoft, has suggested that artificial intelligence could cause the deaths of one billion people in the future. As reports of AI systems potentially escaping human control become more frequent, this warning is raising public concern. David Linthicum, an international expert in artificial intelligence, cloud technologies, and cybersecurity, has shared his views on how likely such a scenario might be.

In an interview with NBC News, Bill Gates said artificial intelligence had become powerful enough to influence events that could result in mass casualties. In his view, the combination of modern AI tools and malicious human intent could pose an extremely serious threat to humanity. Gates also emphasized the need for legislation to regulate the use of artificial intelligence.

To assess how realistic this prediction might be, Tom’s Guide interviewed David Linthicum.

According to the expert, for artificial intelligence to cause one billion deaths, it would need the ability to control critical infrastructure, weapons systems, supply chains, or biological systems. AI systems would also have to be capable of operating without direct human involvement. Such a scenario would require several serious failures to occur simultaneously, including the improper integration of technologies into different systems, failures in management, and the deliberate use of AI for harmful purposes.

Linthicum noted that such a scenario is theoretically possible, but its likelihood in practice is low, provided that people avoid making a series of serious mistakes.

The expert suggested that Bill Gates may have been referring to the possibility of artificial intelligence technologies falling into the hands of terrorist groups or unstable states. Even in that situation, however, there would still be a considerable gap between the intention to use the technology and a catastrophe on a global scale. For this reason, the specialist considers the likelihood of a disaster of this magnitude to be low.

Nevertheless, Linthicum acknowledges that some AI-related risks already exist. He highlighted the need to pay particular attention to data breaches, online fraud, fake videos and audio recordings, and failures in automated systems. Biased algorithmic decisions and the misuse of technology in cyberspace are additional concerns. Some companies may also introduce AI without implementing sufficient safeguards and oversight.

In the expert’s view, one of the main dangers lies in inadequately developed AI solutions connected to the real-world systems of businesses and government agencies. Linthicum said he was especially concerned about systems operated by people who lack the necessary experience and skills to use artificial intelligence safely.

To improve AI safety and maintain control over these systems, the expert recommends several measures. These include limiting system permissions, isolating AI from other systems, and requiring human approval for high-risk operations. It is also essential to log all AI activities, test systems regularly, and avoid giving AI agents excessively broad privileges or unrestricted access to application programming interfaces (APIs).

According to Linthicum, security must be built into a system’s design from the outset. This requires applying the principle of least privilege, monitoring actions, preserving the ability to restore a previous safe state when necessary, and isolating systems. Human trust in a technology depends primarily on how effectively its capabilities can be restricted.

The expert also stressed that relying solely on an emergency shutdown button is not enough. Such a button is useful only if a person recognizes the threat in time and can activate it before damage occurs. A more reliable approach should include multiple layers of control, automatic shutdown mechanisms, limits on the frequency of operations, isolated testing environments, approval procedures, and audit logs. Systems must also be resilient to cyberattacks.

In Linthicum’s view, AI safety must be ensured throughout every stage of a project, not just during the final phase of development. As he noted, an emergency shutdown button may give people a sense of security, but it cannot guarantee safety on its own.

The expert also said that artificial intelligence is lowering the cost of cyberattacks and increasing their frequency. Today, the technology makes it easier to create fraudulent emails, clone people’s voices, and produce fake payment documents. At the same time, AI can help cybersecurity professionals strengthen their defenses.

Despite these developments, specialists must continue following basic security practices. These include multi-factor authentication, verifying received information by phone, and creating backup copies of data. Linthicum emphasized that information that looks or sounds convincing can still be fake.

On the question of whether artificial intelligence could help create biological weapons, the expert takes a more cautious position. He explained that AI can help people acquire biological knowledge, improve workflows, and facilitate preliminary research in this field. However, developing weapons of mass destruction requires specialized laboratory facilities, appropriate materials, and the ability to conduct experiments. These requirements create significant barriers.

Even so, strengthening biological safety measures remains important. The expert recommends screening prompts submitted to AI models, restricting access to dangerous experimental methods, monitoring laboratory orders, and strictly enforcing biological safety standards. If AI is found to be used for dangerous purposes, access should be restricted and the relevant authorities notified.

Linthicum stresses that governments, technology developers, companies, and users must share responsibility for AI safety. Governments should account for risks in legislation and establish liability for harm caused. Developers should build systems that are secure, controllable, and regularly tested. Companies should assess how AI is being used and monitor access to data. Users, in turn, should verify the identities of people they communicate with and protect their accounts.

According to the expert’s overall conclusion, the main threat today comes less from artificial intelligence itself than from people who misuse it or fail to follow essential safety measures.