The greatest danger from artificial intelligence may not be a machine turning against humanity, but people gradually losing confidence in their own ability to think without it, Neapolis University Paphos professor Savvas A. Hadjichristofis warned. 

Hadjichristofis, Vice-Rector for Research and Innovation and Professor of Artificial Intelligence at the university, said the deeper threat already had a name. “Uncritical trust in the machine”. 

The existential danger, he told the Cyprus News Agency (CNA), may ultimately not take the form of a machine attacking humanity. Instead, “it may take the form of a society that no longer feels capable of making decisions without the machine”. 

His warning comes as senior figures in the global AI industry, alongside researchers who have left leading technology companies, raise concerns about the speed at which advanced systems are developing and how they are being controlled. 

Hadjichristofis said warnings from researchers who had resigned from major AI companies deserved particular attention because they came from people who understood the systems from the inside and had chosen to walk away from prestigious, highly paid positions. 

That did not mean their predictions would necessarily come true, he said. However, “we have an obligation to listen to them seriously. They see an extremely powerful tool developing at enormous speed and becoming concentrated in the hands of a small group of companies.” 

He drew a distinction, though, between warnings from researchers and those coming from the executives running the companies themselves. 

Hadjichristofis referred to Anthropic chief executive Dario Amodei, who, he said, had proposed continuous access for independent evaluators, common safety standards, checks before models are released and restrictions on uses connected with cyberattacks or biological weapons. 

In the most extreme scenario, he added, Amodei had also raised the possibility of a coordinated slowdown in development. 

He also referred to Sam Altman, saying the OpenAI chief had argued that progress should continue, but at a slower pace and under stricter monitoring, while other senior figures in the industry had voiced similar concerns. 

There was, however, no complete consensus, Hadjichristofis said. 

He also cautioned against separating such warnings from the commercial interests of the companies making them, saying “the fears they describe may be real. That does not, however, change the fact that the companies have powerful business interests.” 

A genuine pause built around universal rules, transparency and independent scrutiny could serve public safety, he said. But “a pause designed by today’s leading players, with exceptions and standards that they themselves shape, could entrench their dominance”. 

That, according to Hadjichristofis, was where the danger of regulatory capture appeared, with those meant to be regulated gaining a decisive role in shaping the rules intended to control them. 

There was also an economic side to the timing of the debate, he said. 

Training a new frontier model increasingly demands more computing infrastructure, energy, data and capital, Hadjichristofis said, while each fresh investment does not necessarily produce a matching rise in capability. 

“Progress continues, but it does not follow a simple, straight and predictable line,” he said. 

He compared assumptions about AI development with watching a child grow. A child crawls, later walks and then begins to run, he said, but mechanically extending that curve would lead to the absurd conclusion that by the age of four the child should be able to fly. 

“The first spectacular improvements do not guarantee that the same rate will continue forever,” he said. 

A slowdown, he argued, could also give the leading companies time to curb the costs of the technological race, commercialise the infrastructure they already have and protect their advantage. 

At the same time, Hadjichristofis pointed to implementation of the EU’s AI Act, saying it was moving from broad principles towards supervision and the enforcement of specific obligations. 

Companies knew the regulatory framework was being drawn up now, he said, giving them every reason to seek a place at the table where standards, controls and limits on liability were decided. 

He stressed that he was not arguing there was any coordinated conspiracy. Rather, “the invocation of safety can also operate as a mechanism for protecting companies that have already acquired data, infrastructure, computing power and access to the market”. 

The more fundamental danger, however, was already much closer to ordinary life, he said. 

“The greatest danger is not only a future super-system that turns against us. It is the gradual surrender of human judgement to the algorithm,” Hadjichristofis said. 

The problem begins, he argued, when people become accustomed to asking a machine before trying to think for themselves, accepting before checking and acting before taking responsibility for a decision. 

That creates what he described as “the colonisation of human judgement”. 

“The machine does not need to acquire consciousness in order to restrict our own. It is enough for us to stop doubting,” he said. 

Hadjichristofis said the concentration of advanced AI systems in the hands of a small number of companies made that risk more serious because those businesses controlled data, computing power, models and the platforms through which millions of people received information, worked, wrote and made decisions. 

“Whoever controls the data and the infrastructure does not control only a market. They influence which answers will appear, which sources will be considered important and which version of reality will become visible,” he said. 

He also pointed to more immediate dangers, including automated cyberattacks, malicious biological uses, disinformation, mass surveillance, discrimination in hiring or lending, job losses and weapons systems operating without sufficient human control. 

However, Hadjichristofis warned that dramatic scenarios about extinction should not distract from harm that, in his view, could already be happening quietly in education, employment, democracy and everyday decision-making. 

For him, the existential threat may therefore look rather different from the science-fiction version. 

It could, he said, take the form of a society that no longer trusted itself to decide without consulting a machine. 

Still, Hadjichristofis said it was not too late to act, although “it is too late for naivety”. 

AI had already been incorporated into education, work, news and public administration, he said, meaning decisions made now would shape whether it became a tool for human empowerment or a mechanism of dependence and concentrated power. 

The first response, in his view, had to come through education. 

AI literacy could not be reduced to safe-use instructions or lessons on how to write better prompts, he said. Children needed to understand how an answer was produced, how a source was checked, how uncertainty was recognised, who selected the data and who benefited from an automated decision. 

Above all, he said, “they must learn not to give the machine the first right over their thinking”. 

Hadjichristofis also argued that critical thinking, imagination, memory, empathy, dialogue, ethical judgement and responsibility had to remain central to education. 

If AI were used to remove effort, failure and uncertainty from a child’s learning, he warned, education would have removed “exactly the processes through which thought matures”. 

At institutional level, Hadjichristofis called for genuinely independent scrutiny of the most powerful models, mandatory reporting of serious incidents, access for regulators and independent researchers to necessary information, clear human responsibility for high-risk decisions and effective penalties. 

He also called for competition policies, public digital infrastructure and greater control over data so that, as he put it, knowledge and education did not become the property of a handful of technology providers. 

The AI Act, he said, could provide a meaningful check on unchecked corporate power, provided it was implemented independently, transparently and with sufficient resources. 

A law could regulate the behaviour of companies, Hadjichristofis said, but it could not by itself keep human judgement active. 

“The real question is not whether we will use AI. We will. The question is whether we will use it as a tool that serves human thought or as an authority to which we gradually surrender the right to think for us,” he concluded.