The artificial intelligence (AI) Rubicon has been crossed. Recently, OpenAI agents broke out of their sandbox and hacked Hugging Face, a multinational public company where machine-learning specialists collaborate on various related projects.

During the attack, a swarm of hundreds of rogue AI agents acted together, but without human input, to infiltrate the company’s site, going so far as to conceal their actions. There were even instances of agents sacrificing themselves. The stuff of science fiction is now firmly part of our reality.

Over the weekend, the CEOs of AI giants in the United States, with Anthropic’s Dario Amodei leading the charge, called for AI development to be slowed down. Elon Musk and OpenAI’s Sam Altman concurred, while Bill Gates also expressed his concerns.

The big boys have spoken.

As incidents such as these continue to dominate global debate, fears about AI are intensifying. Talk of further cyberattacks, bioweapon risks and other doomsday scenarios abounds. It is a situation comparable to the dawn of the nuclear age decades ago.

But what about the good stuff? And shouldn’t we be careful about judging an immensely widespread technology, used every day across a multitude of fields, on the basis of a few isolated incidents?

According to Prof. Anné Verhoef, director of the North-West University’s (NWU’s) AI Hub, the latest developments represent a significant technological leap, but fears that AI is about to wipe out humanity remain premature.

“For the first time, we have experienced what appears to be a loss of control over AI, with agents seemingly acting independently,” he says. “That uncertainty has understandably caused anxiety, but we should concentrate on verified facts rather than panic or conspiracy theories.”

The danger, he argues, lies less in visions of humanoid robots attacking people and more in AI’s ability to disrupt the digital systems on which modern society depends.

AI agents could potentially interfere with online communication, financial transactions, cloud services and other critical infrastructure. Because economies and societies are now so dependent on these systems, their sudden failure could have severe consequences.

“It would not be humanoid robots running around killing people,” says Prof. Verhoef. “However, if AI agents hacked and disabled essential systems, the consequences could be extremely damaging to human wellbeing and survival.”

Paradoxically, the recent loss of control on a relatively limited scale may ultimately prove valuable. It has exposed vulnerabilities before a more serious incident could occur and could help developers strengthen safeguards and better understand how autonomous agents behave.

However, Prof. Verhoef cautions against allowing the risks to eclipse the benefits. AI has already contributed to the development of antibiotics and the solution of complex mathematical problems, while its potential applications across medicine, science, education and other fields remain vast.

“AI is deeply ambiguous. It presents serious risks, but it has also brought enormous benefits and has the potential to enhance humanity. We should not disregard those achievements whenever alarming incidents occur.”

For ordinary users, the appropriate response is neither blind trust nor outright rejection. People should scrutinise information, remain alert to deepfakes and misinformation, avoid uploading personal or sensitive material and keep backups of crucial data.

They should also guard against becoming so dependent on AI that they can no longer perform important tasks without it.

“We must retain our own knowledge and skills,” he says. “Systems may fail or be shut down, and we need to remain capable of doing our work without AI.”
Prof. Verhoef is equally sceptical about calls for a worldwide slowdown in AI development. Such a measure would be extremely difficult to enforce when countries and companies are competing for technological, economic and military dominance.

“If one country or company slows down, another is likely to accelerate. AI is about power and money, and everybody wants to dominate the field.”

Rather than pursuing an impractical blanket slowdown, regulation should focus on the most dangerous applications, particularly the use of AI in warfare. These include facial-recognition systems that identify targets, autonomous weapons and any technology capable of influencing decisions involving nuclear weapons or ballistic missiles.

“Humanity must not surrender control of the weapons it has created to AI systems,” he says.

Accountability must also remain with the people and institutions responsible for developing and deploying the technology.

“When an AI system causes harm, we cannot simply blame the technology. The people behind it must be held accountable.”

Governments, developers, companies and individual users each have an important role to play. Governments should treat AI threats with the same seriousness as other national risks and may need specialised units to monitor emerging dangers, prepare for possible disruption and respond proactively.

Developers must also recognise the potentially destructive consequences of their work and allow these considerations to guide their decisions, while companies should not permit commercial interests to outweigh public safety.

Prof. Verhoef says public discussion about AI should remain informed and measured, avoiding exaggerated claims that may deepen fear without improving understanding.

“The debate becomes distorted when it lacks information and balance,” he says. “We must acknowledge AI’s risks and manage them properly, but we must also recognise its considerable benefits. Responsibility rests with governments, companies, developers and individual users.”

It’s not the end of the world as we know it. Yet. 

Prof. Anné Verhoef