Yet each side may believe that slowing down unilaterally would make it more vulnerable.
That dynamic was central to the Cold War nuclear arms race.
AI is obviously not identical to nuclear weapons. Nuclear weapons can cause physical destruction on an extraordinary scale through a single detonation, whereas AI is a general-purpose technology with both civilian and military applications.
But the comparison becomes relevant when considering strategic competition.
The more powerful AI becomes, the more governments may see it as a source of military advantage.
And once military advantage becomes the objective, pressure to move faster can conflict with the need for safety, testing and international regulation.
The Speed Problem
Cameron's most unsettling observation concerns speed.
He described a hypothetical combat environment in which AI systems could fight at a pace that humans could no longer meaningfully intervene in, leaving little opportunity for de-escalation.
That is a critical concern.
Humans are not particularly good at making complex decisions under extreme time pressure.
Machines, however, can process information and execute predefined actions extremely quickly.
If two military AI systems are responding to one another at machine speed, the traditional human decision-making loop could become dangerously compressed.
Imagine a hypothetical sequence.
One system detects what it believes is an incoming threat.
It recommends or initiates a response.
The opposing system detects that response.
It interprets the action as hostile.
It prepares a countermeasure.
The first system detects the countermeasure and escalates again.
At every stage, the machines could technically be responding to data.
Yet the original interpretation could be wrong.
A sensor could malfunction.
An algorithm could misclassify an object.
An intelligence report could be incomplete.
An adversary could deliberately feed misleading information into the system.
Or an AI could simply make a mistake.
Humans make mistakes constantly.
The difference is that humans can sometimes recognize their mistakes and stop.
The danger with automated systems is that a mistake could become an input for the next machine decision.
That is how a chain reaction could theoretically develop without anyone intending it.
The United Nations Is Raising Similar Concerns
Cameron's warning is not the only voice raising alarms.
The United Nations has increasingly focused on the military implications of artificial intelligence.
In 2024, the UN Office for Disarmament Affairs published an extensive study examining the governance of AI in the military domain. The publication noted that AI could create opportunities for militaries but also amplify existing risks and generate new unintended consequences. It specifically highlighted the concern that technological development may be moving faster than the creation of effective international guardrails.
That is significant because it shifts the conversation away from science fiction.
The issue is no longer simply whether Hollywood's predictions are accurate.
Governments and international institutions are now asking practical questions.
Who is responsible when an AI-enabled military system makes a deadly mistake?
How should such systems be tested?
What level of human supervision is necessary?
Should certain military applications of AI be prohibited?
How can countries verify that rivals are following agreed rules?
What happens when an AI system behaves unpredictably?
And perhaps the most important question of all:
Who has the final authority to use lethal force?
NATO Is Also Preparing for the AI Era
NATO provides another important indication of how seriously governments are taking the military implications of artificial intelligence.
In July 2024, NATO released a revised Artificial Intelligence Strategy designed to accelerate AI adoption while emphasizing responsible use.
The alliance acknowledged that AI could significantly enhance military capabilities but also identified risks associated with reliability, accountability, adversarial use and human-machine interaction.
NATO's principles include responsibility and accountability, explainability and traceability, reliability and governability.
The principle of governability is particularly relevant to Cameron's concerns.
NATO states that AI applications should permit appropriate human-machine interaction, allow unintended consequences to be detected and avoided, and provide mechanisms such as disengagement or deactivation when systems behave in unintended ways.
In other words, one of the world's most powerful military alliances is explicitly considering how to ensure that humans remain capable of controlling AI-enabled systems.
That is not Skynet.
But it is a recognition that advanced military AI creates a new category of technological risk.
Human Control Is Becoming the Central Issue
The phrase "human in the loop" has become increasingly important in discussions about autonomous weapons.
The basic principle is simple: humans should retain meaningful control over decisions involving lethal force.
But defining "meaningful control" is much harder than it sounds.