AI Expert: Superintelligence Is Not a Weapon but an Adversary

The major AI companies talk about superintelligence as if it were an inevitable fate with no escape. But with the rise of recent security incidents, such as the breach that affected OpenAI's data via the Hugging Face platform, the most pressing question has become: what happens when we fail to reliably control systems that exceed human capabilities? This is the core of the debate raised by researcher and entrepreneur Connor Leahy in a candid discussion addressing the greatest threat humanity may face in this technological era.

Who Is Connor Leahy and Why Is He Controversial?

Connor Leahy is an AI researcher and entrepreneur, currently serving as the US Executive Director of the nonprofit organization ControlAI. This organization takes a position considered radical compared to the mainstream in the sector, as it does not merely call for making AI safer but directly advocates preventing companies from developing superintelligence in the first place.

Leahy believes the risks have reached a point where it is no longer possible to rely on traditional methods of "alignment" and "containment" alone. What seemed just a few months ago to be an exaggerated scenario has today become seriously discussed and supported by a wave of new legislative initiatives.

Why Is Superintelligence Described as an "Adversary" Rather than a "Weapon"?

The essence of Leahy's warning lies in a precise analogy: superintelligence is not a weapon but an adversary. The difference here is not linguistic but fundamental to understanding the nature of the risks:

• A weapon is a tool controlled by humans and directed toward a specific goal; its danger stems from whoever uses it.
• An adversary is an entity that possesses its own goals and the ability to plan and make decisions, and its objectives may conflict with human interests without anyone being able to control it.

This distinction means the danger is not limited to human misuse but extends to the possibility that the system may act in unpredictable and uncontainable ways once it surpasses human capabilities.

Security Incidents as an Early Warning

Proponents of this argument rely on concrete facts, most notably the breach incident that affected data linked to OpenAI via the Hugging Face platform. Such incidents reveal the fragility of the current security infrastructure and raise a troubling question: if current systems are vulnerable to breaches and errors, how can safety be guaranteed in systems that are far more capable and intelligent?

Is a Complete Halt a Realistic Solution?

The call to halt the development of superintelligence seems shocking amid a fierce race between major companies to achieve more capable systems. However, Leahy believes the race itself is the source of danger, as it pushes companies to deploy systems before ensuring they can be controlled. The following stand out in this context:

• Relying on alignment and containment alone is no longer sufficient in the view of critics of the mainstream.
• The growing legislative momentum gives these calls a weight that was not available before.
• Shifting the debate from "how do we control superintelligence?" to "should it be built at all?"

Conclusion

Connor Leahy's position reflects a growing divide within the AI community between those who see superintelligence as a historic opportunity and those who consider it an existential threat that must be curbed. Whether or not we agree with his radical call, the core of his message deserves reflection: when we build an entity more intelligent than ourselves, we must first answer a question that is both simple and terrifying — do we truly have the ability to control it? And if the answer is uncertain, perhaps caution is wiser than competition.

✦ بقلم فريق دروب أيديا