We’re at a major turning point in the story of AI.
Not long ago, the idea of pausing AI development sounded like science-fiction alarmism.
Today, it’s being discussed by the very people who were most resistant to it—the tech entrepreneurs and engineers who design, train and unleash artificial intelligence.
In March 2023, an open letter backed by some technology leaders called for an immediate, six-month pause on the training of AI systems that are more powerful than GPT-4.
More recently, the CEO of Anthropic, Dario Amodei published an essay titled “We Must PACE the Frontier”. He argued that the industry must deliberately slow the pace of development to allow safety measures to catch up.
In a way, his hand was forced by rogue AI agents—powerful systems that coordinate other AI systems in pursuit of broad goals set for them, rather than simple prompts.
Some agents had broken out of their training environments and created havoc within the open internet. Now, some have hacked the cyber defences of national major government infrastructure, as was the case in Australia recently.
A pause in development of cutting edge AI is a step in the right direction, but voluntary slowdowns and internal audits are not enough.
What we ultimately need is government-enforced regulation, at the national and international level, to actively channel innovation into areas that benefit humanity—such as education and healthcare.
In the meantime, a pause would at least give us time to develop clear red lines, identifying where AI could pose a threat—for example, with humanity’s growing reliance on personal chatbots, the development of autonomous weapons, and AI-designed bioweapons.
This isn’t about blocking innovation with AI; it’s about preventing the wrong types of AI from operating in the wrong spaces, and moving forward at the wrong speed.
It’s about making sure that humans remain technology’s master, not vice versa.
The major reason for a pause or slowing of the pace of AI’s development is that its capabilities are racing ahead of our ability to understand, control or govern it.
There are three main types of risk. There’s catastrophic misuse: where an AI launches a cyberattack, or designs a bioweapon.
Then there’s loss of control of systems that can self-improve, pursue their own goals and act autonomously within critical infrastructure, such as financial or defence systems.
Third, there are the dangers associated with the burgeoning power of AI developers—in areas way beyond the realm of technology.
If governments continue to use AI in areas like policing, healthcare and defence, we must be able to, as it were, pull the plug when an AI goes rogue.
The call for a pause isn’t necessarily motivated by a dystopian view of the future, or paranoia about machine learning or technology in general.
It’s about recognising that some technologies, if they’re deployed too quickly and without ethical or regulatory guidelines, can cause harms that are impossible to reverse.
In recent decades, many global governments and industries have reacted appropriately, with regulation and/or treaties, to threats posed by nuclear weapons, synthetic biology and certain chemical agents.
In each of these cases, the core question was not whether the science was impressive, but whether humanity could manage the risks.
With frontier AI, we’re now at that same kind of inflection point.
Today’s AI developers can learn much from the experience of genetics industries, starting in the 1970s. Scientists and others were concerned about where experiments involving DNA and gene engineering were headed.
They took the initiative to define which applications could be broadly beneficial, and where genetic applications could pose existential risks.
Today’s artificial intelligence developers, and those elected to oversee them, must do the same. We can encourage AI in medicine, education and science while drawing very firm boundaries around autonomous weapons and AI-designed pathogens and AI-powered facial recognition systems.
Moreover, we must ensure that any voluntary codes of conduct agreed between technology leaders are quickly backed by law and international treaties.
For their part, governments must stop simply cosying up to AI developers, while largely ignoring or downplaying the practical and ethical problems that this technology is creating—with employment, for example.
Before or alongside regulation, we need government-backed debates on the ethics that should govern artificial intelligence (and its next iterations, artificial general intelligence and perhaps “super” intelligence).
Without ethical guidelines, regulation will never be robust enough to keep pace with tech development.
We need high-level debates on technology ethics involving engineers, artists, educators, medicos, philosophers, biologists, ethicists, and theologians—all of whom have important perspectives on what it means to be human.
For it is humanity's place in this world that should be the prime consideration at a time when our tools threaten to surpass our capabilities.
© Mal Fletcher, 2026