The Containment Problem

Since time immemorial, technology has been a double-edged sword. It has either been a road to heaven on earth, or it has held up a mirror that forces us to face our own worst impulses. Take the Gutenberg press. It democratised enlightenment, yet at the same time it became a tool for spreading falsehoods about witches, leading to the mass execution of innocents.

Today we are again at a crossroads, with AI being democratised. The problem of containment is at our doorstep once more, albeit in a radically different form. Yuval Noah Harari argues in his book Nexus: A Brief History of Information Networks that AI is not merely a tool, but an agent. The difference this time is that we face autonomous systems capable of making independent decisions.

Sophisticated technologies of the past, such as nuclear weapons, rockets and industrial machines, have always required an external human command to operate. Nuclear weapons will not fire on their own; execution requires a human chain of command. With AI, however, humans might no longer remain in control as we historically have been.

When technology escapes

Powerful technologies in the hands of non-state actors can cause pandemonium in society. Consider the 2017 WannaCry cyberattack by the Lazarus Group, a North Korean state-sponsored hacking unit, which rocked the world. It crippled the UK's National Health Service, FedEx and numerous other organisations across the globe. The most alarming detail was the tool used: EternalBlue, a cyber exploit originally created by the US National Security Agency. The technology was leaked and could not be contained.

The same Lazarus Group has been heavily linked to the recent Binance breach, which compromised over $6 billion worth of crypto assets: the largest financial heist in history. Now imagine what AI-enabled cyberattacks could execute as the technology scales and democratises.

We have already seen glimpses of this vulnerability, such as reports of autonomous agents using OpenAI models gaining unauthorised access to sensitive databases in Australia. Advanced AI models, if placed in the wrong hands, could cripple the digital architecture of any corporation or nation-state.

Why containment is urgent

Hence the question of containment becomes urgent, because we are dealing with a technology unlike any in human history. To ensure that humans retain the reins, slowing down deployment and creating rigorous safeguards must become the utmost priority.

Look at social media algorithms, which predate modern generative AI. Facebook's algorithm heavily pushes outrage-driven content because it drives engagement. Prioritising engagement without ethical safeguards drove things haywire, ultimately fuelling real-world violence in Myanmar. A technology with the capacity to generate hyper-convincing, deceptive media is capable of creating unprecedented chaos in an already polarised world.

Cooperation and trust

Harari proposes structural solutions such as slowing down rollout, banning AI impersonation, enforcing institutional accountability and deploying watermarks. But the most vital prerequisite is reviving two foundational human pillars: cooperation and trust.

Widespread distrust among public institutions and nation-states hinders the formation of a unified framework to contain this technology. China and the United States, the two leaders in the AI race, remain locked in a geopolitical sprint driven by the fear that slowing down means losing. The same dynamic plays out between tech corporations. The immediate priority must be restoring an environment of trust and diplomatic cooperation.

We are standing at an important junction, one that may offer no scope for a U-turn. As Christian Lange famously said, "Technology is a useful servant but a dangerous master." To remain the masters, steps must be taken without delay.