Where machines fail, will human instinct suffice?

PHOTO: ACatInaBox on Unsplash

This year began with the Doomsday Clock moving ten seconds closer to midnight, a  symbolic clock designed by scientists to represent how close humanity is to destroying the world. This placed us closer to catastrophe than at any point since the clock’s establishment in 1947. Merely a month later, The New Start officially expired, the last remaining nuclear arms control agreement between the United States and Russia signed in 2010 to cap the amount of nuclear warheads of each state. 

In that same month, a report from the Stockholm International Peace Research Institute (SIPRI) found increasing signs that China, India and the United States are integrating AI into nuclear and nuclear-related systems. Emotion is commonly associated with war and destruction, yet fear and empathy have arguably been the only things which have saved our planet from annihilation — emotions that artificial intelligence (AI) cannot feel.

This May, 130 countries attended the Review Conference of the Nuclear Non-Proliferation Treaty, however diplomatic deadlock over the current war in Iran and great power politics meant that they failed to produce a final document. This means that the treaty has failed to reach consensus for over fifteen years. 

The mix of declining nuclear arms controls and rising AI usage in defence systems raises the question: what happens when humans are distanced from the consequences of their actions? 

We wouldn’t be here without human instinct

PHOTO: President John F Kennedy meets with members of the Executive Committee of the National Security Council regarding the missile crisis in Cuba, The White House on Wikimedia Commons

The battle between human judgement and machines is not only seen in Star Wars: Attack of The Clones. In the Cold-war, science-fiction merged with reality during the Cuban missile crisis. 

On 27 October 1962, US forces began dropping non-lethal depth charges, an underwater signal to indicate that they knew there was a Soviet submarine. However, the soviet crew onboard the submarine had no contact with the outside world and feared the onslaught of WW3. 

On board and unknown to the US they had a nuclear weapon and the permission to launch it without Moscow approval. Two of the vessel’s senior officers wanted to launch the nuke. However, the last officer refused and the torpedo was never fired. Luck and gut instinct saved a majority of those born before October 1962. 

Evidently, where machines have let us down, humans have stepped up to the challenge. Monash University academic, Ben Zala, who specialises in nuclear politics and security studies, argues that an AI system would struggle to replicate this kind of cautious, context-dependent judgement. 

The objective of using AI is to speed up intelligence analysis. Therefore, Zala argues that even if humans retain the final decision to launch nuclear arsenal, this decision is informed by data analysed or collected by AI, even if that data itself is ambiguous. 

The idea of advanced technology to defend against nuclear attack is nothing new. In 1983 Ronald Reagan declared the Strategic Defensive initiative, nicknamed “Star Wars” by opposition. 

Yet, unlike the Death Star, the initiative never actually got off the ground. Last year Trump announced the Golden Dome defence project, with many parallels drawn between the projects. Both the Death Star and Golden Dome projects made global headlines and are hugely expensive and challenging to operate. 

The Golden Dome project intends to protect the entirety of the country from missile attacks and be completed by the end of Trump’s four year term. In March, this year the Wall Street Journal reported that Palantir and Anduril are developing the Golden Dome’s software, companies that specialise in integrating artificial intelligence. 

So … what is AI used for in nuclear systems?

According to the SIPRI report, when it comes to China’s nuclear forces and its fleet of delivery systems — which include the Dongfeng series of intercontinental, intermediate-range, medium-range and short-range ballistic missiles — AI has the potential to enhance precision guidance and target selection and navigation. In the air, China possesses and exports a vast array of uncrewed systems that could have impacts on nuclear forces in terms of surveillance, disruption or even attack. 

China is in the process of becoming a leader in automated strategic non-nuclear weapons, demonstrated by their 2015 White Paper, which outlines that China will “improve strategic early warning command” to ensure the survivability of their state. 

In February 2025 the Director of India’s Defence Research and Development Laboratory stated that the laboratory is using AI to streamline missile production, including its Agni, Prithvi and Akash missile variants which are used to carry nuclear warheads. The US has also made breakthroughs in developing AI agents and testing uncrewed systems that can be deployed in tandem with and in support of nuclear platforms so that they may fire when ready. 

China’s advancing development of automated and autonomous strategic weapons, including AI-enabled early warning and unmanned systems, threatens the global nuclear order by undermining traditional deterrence mechanisms and increasing the risk of pre-emptive strikes.

PHOTO:: The Dong-Feng 26 ballistic missile, IceUnshattered on Wikimedia Commons

AI impacting traditional deterrence in the third nuclear age

Whilst the government and political strategists are not saying that AI is taking charge of nuclear decisions, AI is being integrated into nuclear systems that inform human decision making. ICAN believes that the use of AI in the military domain unveils the shortcomings of nuclear deterrence. 

According to the Carnegie Council: “Nuclear deterrence refers to a principle in international relations where the retaliatory potential and destructive force of nuclear weapons prevents nations from launching a nuclear attack.”

AI is designed for efficiency, yet as reaction times shrink, the risk of catastrophic miscalculation grows. These systems are also vulnerable to hacking and other interference. Therefore AI magnifies the flawed assumption that deterrence is defined by rational control and perception. ICAN states that “Algorithms cannot be coerced, bluffed, or frightened into restraint”. Essentially, nuclear deterrence and mutually assured destruction depends upon fear which AI cannot feel. 

AI is also thought to challenge the stability-instability paradox, the theory that nuclear weapons decrease likelihood of major wars fought between nuclear powers but increase the risk of smaller proxy wars.

Senior Lecturer in Strategic Studies at the University of Aberdeen, James Johnson argues that increasing the integration of AI technology with nuclear command and communication systems enhances states’ early warning systems,early threat detection, nuclear communications and intelligence analysis. However, according to Johnson, while this may sound stabilising, it in turn raises the threshold below which a nuclear first strike is successful. Essentially, leaders may become more confident that they can control escalation and push the envelope further in crisis due to an increased trust via AI. 

Leaders must also interpret AI recommendations, cyber attacks and misinformation in a quicker amount of time if AI says there is an impending threat, potentially raising the likelihood of hasty decision-making.

“Improving the brakes on a car would not make playing [the game of] Chicken safer, but only allow them to drive faster to produce the same level of intimidation”

– Robert Jervis, Columbia Professor of International and Public Affairs

PHOTO: Wikimedia Commons 

Johnson hypothesises that in a scenario where China and the US deploy AI decision making technology into their networks, the likelihood of unintentional atomic detonation increases.

The Third nuclear age, the current era of nuclear competition, is defined by multipolarity and the rivalry between the US and China and Russia. 

While China has not placed nuclear warheads directly in the South China Sea, it has militarised 20 outposts in the Paracel and Spratly Islands. 

Given that the Pentagon reported that Beijing would acquire the military capabilities to forcefully take Taiwan by 2027, the ability of AI to potentially destabilise traditional modes of deterrence comes as a wake up call for policymakers. 

In July 2026, China test-fired a nuclear-capable intercontinental ballistic missile (ICBM) from a submarine in the Pacific Ocean. This test came just days after Australia and Fiji made a new defence alliance that commits to come to each other’s aid.

However, Defence Minister Richard Marles said he did not believe the test was designed as retaliation for the defence treaty struck between Australia and Fiji, as reported by the Sydney Morning Herald. 

While there may be little correlation between the test and the defence treaty, the test may have sent shockwaves through Australian policymakers, hitting a little too close to home. 

Data centres Down Under

PHOTO: Virginia Tech Data Centre, Cristopher Bowns on Wikimedia Commons

There are currently over 290 data centres in Australia with an additional 100 facilities in development, largely in Sydney and Melbourne.

In July 2026 Anthony Albanese told Australians that “AI is a critical – and urgent – innovation priority for our defence force and security agencies”. 

The federal government announced in 2024 that three secured data centres would be built in undisclosed locations to support an Amazon secret data cloud with military and intelligence information.

Defence Australia says that “Human judgement and accountability is central to lawful, legitimate and responsible use of AI, and designated Defence personnel (Accountable Officers) will always be accountable for its use, decisions and outcomes”. 

Drone strikes on data centres in Bahrain this year amid the middle east conflict show that modern conflict depends on data rather than just ammunition. Defence expert Ian Langford argues that this presents an opportunity for Australia to be the regional hub of data centres due to its physical distance from conflict zones and trusted technology environment. 

The increase in the accessibility and use of AI across Australia also poses significant cybersecurity threats. AI presents an opportunity to improve detection of threats and vulnerabilities and reduce reliance on repetitive tasks. Simultaneously, as AI becomes more accessible, malicious actors will continue to increase the speed, scale and impact of malicious cyber operations. 

As 81 per cent of Australian organisations have experienced an AI related security incident or identified an AI vulnerability in their system, the AI dilemma doesn’t seem as though it will be leaving cybersecurity discourse any time soon. 

Unmanned aerial vehicles: AI drones inducing bloodshed

PHOTO: Paul Brennan on Wikimedia Commons

Autonomous battle droids may have once belonged only to science fiction however nowadays, nuclear weapons aren’t the only weapons where humans are being taken out of the loop. In Ukraine, AI drones have been used to strike Russian targets at greater distances with increased accuracy with the AI-enabled Hornet system. Similarly, Israel has identified 37,000 Hamas targets through AI applications: the Lavender and Gospel systems. However, it is not just governments that are capitalising upon this technology. 

ACLED in 2025 reported that 469 armed groups, including rebel groups, militias, gangs, and transnational cartels used drones in the past five years. This was just under 50 times the reported amount of groups using drones in 2020. Terrorist groups like ISWAP or JMIN are increasingly using drone strikes in Niger, Nigeria, Burkina Faso and Mali on civilians. 

The new shift is further being accelerated by offline AI tools that can help drones avoid traditional detection and jamming methods. They are also using these tools to generate training material and press releases.  

Who bears the blame? 

Beyond drones, AI has been used to inform conventional  attacks including the current conflict in the Middle East. The US military reportedly used Claude, Anthropic’s AI model, to inform its attack on Iran, 

In conventional warfare, man pulls the trigger; when AI is used the weight of that decision shifts, and so too does blame. The usage of AI in informing conflict decisions and drones strikes the question, might it be easier to blame AI companies for faults of civilian casualties instead of taking accountability? 

The Doomsday Clock is still ticking. The question remains whether we let the machines decide how fast that is amid the battle with artificial and armageddon.

If you want to hear more about nuclear weapons, arms control and the Middle East you can listen to season 6 episode 1 on ATLAS Spotify or YouTube featuring Security Studies and nuclear politics Academic Ben Zala. ATLAS is a Monash International Affairs weekly Video-Podcast.