An Israeli scientist calls on governments to link AI agents and systems to concrete persons and hold them accountable when AI goes rogue.
Israeli historian and philosopher Yuval Noah Harari has made an interesting observation about artificial intelligence (AI) during a recent interview with Czech Television’s Hyde Park Civilization show by comparing AI agents to immigrants.
Following are several important ideas worth transcribing from that interview. I reproduced those ideas as close as I could, preserving the language and boredom of facts.
Idea 1: AI immigrants
The professor of the Hebrew University of Jerusalem believes the cross-border spread of AI systems appears in many ways similarly to human migration.
Developed largely in the United States and China, they can transform labor markets, cultures and political relationships without immigration controls or meaningful input from the societies adopting them. They may displace workers in journalism, accounting, banking, law and healthcare while importing the priorities of the corporations and governments controlling them.
AI agents and systems enter new countries without border control or visas, and they don’t become loyal to the new jurisdictions – they remain loyal to their creators and owners, Harari noted.
The danger becomes acute especially when AI enters critical infrastructure: a country whose airports, hospitals and military depend on foreign systems could face enormous pressure if a provider remotely withdraws access, the researcher said, pointing to Ukraine’s dependence on Starlink as an illustration of the leverage conferred by control over essential technology.
As AI becomes embedded across economies and public administrations, such dependence could become permanent: rapidly evolving models require continuous updates, leaving customers reliant on their suppliers.
For smaller European countries, independent alternatives may be prohibitively expensive. The European Union could achieve the necessary scale by pooling resources and building its own technological capabilities.
Political fragmentation, Harari argues, weakens Europe’s ability to regulate powerful technology companies and protect its sovereignty. The contest is therefore about more than building the most intelligent model. It is about controlling the infrastructure on which societies depend — and retaining the ability to make independent decisions.
Idea 2: Demolition of trust
Harari also warns that AI threatens the shared and generally-accepted understanding of reality. It can generate convincing images, voices and videos of people saying and doing things they never said or did.
As an example, he described watching a fabricated video of himself discussing AI and needing about 90 seconds to become almost convinced it was fake. His team had to investigate, while even his sisters, who had known him for decades, could not confidently identify a much longer version as fabricated.
The deeper problem, in his opinion, is trust, which is based on truth. But truth is expensive: it requires time, expertise, investigation and a willingness to confront inconvenient facts. Lies and fantasies, on the other hand, are cheaper to produce and can be tailored to confirm prejudices or exploit fears.
Professional journalists, universities, and other institutions provide methods for checking claims and challenging evidence. When those procedures are abandoned, information becomes an avalanche in which people struggle to distinguish reality from fabrication.
Harari connects this confusion to authoritarian politics. Dictators do not need to convince everyone of their propaganda; they may succeed simply by destroying confidence in every source of information. That’s what U.S. President Donald Trump does systematically. That’s what Russia’s Vladimir Putin has done too.
When citizens cannot agree on basic facts, collective action and public accountability become difficult, the scientist warned. AI could accelerate this process by making fabricated evidence abundant, inexpensive and increasingly convincing.
And this is what is exactly happening.
Idea 3: Information labelling
Harari traces part of this crisis to algorithms that determine what billions of people read and watch. Unlike traditional newspapers, where human editors select and prioritize stories, social platforms increasingly rely on algorithms designed to maximize engagement and advertising revenue.
These systems exploit a powerful emotional trigger: outrage. An ordinary account of government bond yields or fiscal changes – which may be important news for economy – attracts little attention compared with a shocking allegation or a frightening story, even if the latter is misleading or false.
According to Harari, this exploits an evolutionary survival mechanism.
“Our ancestors benefited from noticing threats quickly; someone who spotted a snake on the savannah was more likely to survive. Social media turns that adaptive instinct into a vulnerability by repeatedly exposing users to content designed to provoke fear, anger, envy and indignation,” he explained.
The result is continuous agitation that distorts perceptions and intensifies hostility between various groups.
The problem, however, extends beyond individual falsehoods: the entire information environment can become organized around emotional manipulation rather than truth, contributing to social fragmentation and potentially to societal collapse.
Harari proposes strengthening public-service media, supporting professional journalism and restoring human editorial judgment based on accuracy, importance and responsibility rather than clicks.
He also suggests labelling content according to its emotional characteristics, much as food packaging identifies salt, sugar and fat. Users could recognize when their information diet consists largely of outrage, fear or greed.
People increasingly monitor what they eat while neglecting what they feed their minds, he repeated a modern aphorism.
Idea 4: Attribution of responsibility
Yuval Harari strongly objects to the claim that the rise of superintelligent AI is inevitable and that governments and citizens have no meaningful choices left.
He uses the example of a rocket that has a significant chance of exploding. Even if someone argues that the passengers are already aboard and must therefore accept the risk, Harari questions why society should treat that situation as an excuse for surrendering control.
The language of inevitability, he argues, removes technology from democratic debate. Politics exists precisely because societies can choose between different courses of action, establish limits and change direction when circumstances demand it.
Should all AI development stop? Not necessarily, he says. What must be done is to pressure governments to regulate dangerous practices, to demand greater investment in safety and to prohibit systems that engage in risky forms of autonomous development.
Among such practices the scientist identifies AI systems pretending to be human beings and systems developing successive generations of AI without meaningful human supervision or intervention.
The problem, in his view, is that the companies developing AI have powerful incentives to prevent regulation. They benefit from presenting their products as controllable and safe when attracting investors, yet distance themselves from responsibility when those systems cause harm.
Harari argues that AI responsibility must remain attached to identifiable human beings and organizations. A company should not be able to claim that it controls an AI system when seeking investment, then argue that it bears no liability because that system had acted independently.
All AI agents and systems must be linked to concrete persons and companies – their creators and owners, he stressed.
Idea 5: Democracy is worth defending
For Harari, democracy’s defining strength is its capacity to correct mistakes. Elections allow citizens to replace leaders, but independent courts, a free press and other checks and balances are also necessary to prevent those in power from manipulating the system to remain in office.
Dictatorships, on the other hand, lack effective mechanisms for correcting leaders’ errors. When power is concentrated and criticism suppressed, disastrous policies can persist because no institution can compel reconsideration.
The same challenge applies to AI: systems too complex for humans to understand may evade meaningful oversight; companies may escape responsibility, while governments weakened or captured by commercial interests may fail to impose the necessary limits.
If the technological oligarchy goes through and takes control over governments and institutions, we’ll see a new form of imperial power, unchecked and unaccountable to anybody.
“No system should become so powerful that its errors can no longer be questioned or its decisions held accountable,” he stated.
The next five years, he suggests, could radically change employment, relationships and everyday life. Preparing for this future requires more than defending existing arrangements: societies must adapt while retaining their ability to choose.
Governments must therefore slow the development of dangerous capabilities, invest in safety and establish clear enforceable rules. Humanity should retain the power to decide which risks to accept, which practices to prohibit and which institutions to strengthen.
“The greatest danger is surrendering the belief that such choices remain possible,” Harari noted.
Yuval Noah Harari is the author of the bestsellers Sapiens: “A Brief History of Humankind”, “Homo Deus: A Brief History of Tomorrow”, and “21 Lessons for the 21st Century”. His books sold around 55 million copies in 65 languages.
He is also the co-founder of Sapienship, a social-impact company focused on research, education and storytelling, and the board director at LawZero, an AI safety organization.
***
This story is free. Follow me for more. If you feel you can contribute to my efforts, consider becoming a supporter or a member.