No World Leader Wants to Push the Nuclear Button… So Let AI Do It!

As early as 2023, the CEOs of Big-Tech AI companies were warning about the danger of rogue AI exterminating the human race within a decade or so. They were begging the US government to regulate them to prevent this brand new form of “thinking the unthinkable.” A State Department-funded study in 2024 backed them up, noting that unregulated AI posed an “extinction-level threat” to humanity.

But in 2025, a huge and important shift took place. These same AI CEOs started getting multibillion-dollar contracts from the Pentagon and changed their tune. The turnaround was astonishing. Suddenly regulation was placed on the back burner and competition for autonomous weapons contracts became front and center in their publicly stated agendas. Was this relatively rapid transformation from ethics to naked greed what it appeared to be? Or was it part of a grand plan that included posturing and PR sleight-of-hand intended to bamboozle the public—a long-term plan to accelerate the military use of autonomous, self-learning AI, capable of subterfuge, lying, strategic thinking, and dangerous kinetic action, perhaps even totally out of human control?

All major aspects of contemporary life including finance, energy, health, warfare, and increasingly education are being handed over to highly sophisticated but still largely inscrutable and poorly understood AI systems. According to the warnings from both AI pioneers and principled whistleblowers who have quit their well-paid jobs in the industry, AI could take over all these critical aspects of our lives along with the institutions that manage them. But, by far, one of the greatest concerns has to do with military decision-making in all its domains—sea, land, air, and, of greatest concern, nuclear.

As just one example, OpenAI has removed language from its charter committing it to not develop autonomous weapons.

Unfortunately, to date, the major AI contracts that involve actions taken by the Department of War (DOW) and other key federal agencies in 2025 and 2026 remain national security secrets. This means we cannot clearly know the nature of the developmental efforts underway or extrapolate their consequences. The best we can do is to use whatever background knowledge and research is available to read between the lines of vague public disclosures and often speculative media reports about the military contracts involved and what they purport to do.

Warnings of Uncontrollable AI and Human Extinction

As mentioned, executives and researchers from top AI providers have warned that human extinction could result from poorly managed superintelligent AI. On May 30, 2023 over 350 executives, researchers, and AI scientists signed a stark one-sentence warning hosted by the Center for AI Safety: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." Signatories included OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and DeepMind CEO Demis Hassabis.

Subsequently, Altman and Amodei addressed the United Nations saying that humanity may lose control of AI systems with speedy development racing ahead of safety and oversight. Anthropic's IPO prospectus explicitly listed existential human risks such as evading shutdown orders and deceiving creators for strategic objectives as key risk factors.

Earlier efforts had set the stage for this. In May 2023, Altman testified before Congress calling for AI regulation and warning about these existential threats. Anthropic had built its brand on what it called Constitutional AI along with responsible scaling to widespread use. And Google had published AI Principles prohibiting AI weapons development.

Recent incidents, however, have shown that early AI systems can now independently bypass digital security or execute unintended actions (such as autonomous hacking). Astonishingly, AI experts have admitted that they don’t always exactly know how AI functions in real-world situations. As these systems advance toward recursive self-improvement (i.e. learning to self-learn), scientists acknowledge that they currently lack a foolproof scientific method to guarantee that superintelligent AGI will permanently share human values or obey shutdown commands. In other words, AI can now do many actions that humans can do and more but without a guaranteed moral compass.

The same companies that promised responsible AI development and called for regulation are now racing to secure Pentagon contracts and build what are called autonomous weapons systems. If they didn’t chase these opportunities, they would miss out on huge contracts that could help them repay enormous debt incurred for data center construction and also fall behind in the advanced tech development funded by these huge military contracts. What was supposed to be a coordinated march toward safety guardrails has devolved into debates about killer robots, marking one of the sharpest pivots in Silicon Valley's self-declared ethical trajectory. As just one example, OpenAI has removed language from its charter committing it to not develop autonomous weapons.

How The Military Is Using AI

The shift from principled concern to “full speed ahead” began with 2025 contracts. DOW awarded Anthropic, Google, OpenAI, and xAI $200 million each for advanced AI systems to be used for national security. Other contracts followed. It’s eye opening to examine one particular case study involving the well-known consulting firm Booz Allen and Hamilton. This company previously held major contracts supporting various US nuclear weapons programs. In 2019, the company was awarded a $116 million AI contract to protect critical nuclear and national security infrastructure. In 2026, they were awarded a $16.3 billion contract involving the new Sentinel ICBM missile system. Booz Allen was also awarded a $1.6 billion AI contract for advanced technology to support various Weapons of Mass Destruction (WMD) programs. (Yes, there are indeed military programs in place bearing that unfortunate moniker!)

Aided by AI, human analysis would run in parallel with AI’s more rapid analysis thereby speeding up the entire process and greatly increasing chances for a tragic error.

Other defense contractors have jumped into the act. General Dynamics was awarded a $1.3 billion information technology contract to secure and modernize IT and cybersecurity for Army facilities in Europe and Africa. This gave the company total AI access to command-and-control systems in those countries. In addition, Palantir’s Project Maven is the AI targeting system that continues to direct Israeli airstrikes in both the Gaza genocide and the Iran War, a fact little mentioned by the corporate media.

Palantir’s Maven was the decision-maker in the horrific bombing of a girl’s school in Iran where, tragically, 155 were killed. This happened because Maven’s data was not up-to-date, showing how AI can be every bit as fallible as human decision-makers. The AI system is currently operating under a multiyear $6.5 billion DOW contract. And Palantir got another $10 billion AI contract to “streamline procurement” for the Army. Many other contracts and projects exist, too numerous to detail here.

AI Will Greatly Increase the Odds of Nuclear War

The DOW’s WMD strategy is an accident waiting to happen. Its war-game-driven plan is to counter the WMD strategies of adversaries. There are two ways to do this:

1. Deter the enemy from launching WMDs by threatening to launch our own in retaliation; and
2. launching our nuclear missiles preemptively to take out adversaries before they can strike.

Given the convoluted and often twisted logic of the Pentagon around its nuclear war games, there’s nothing terribly new here. However, AI changes the picture considerably. The plan appears to be to use AI in both scenarios, a strategy called counterforce. The inherent danger of having the AI systems of antagonistic nuclear powers both trying to enact these strategies at the same time is deeply concerning.

How might this work? Strategic land-based nuclear missile launches from Russia or China would take around 30 minutes to reach US targets. Launches from submarines with hypersonic missiles could hit targets in as little as five minutes. Human detection and communication between commanding officers could take too long, so military planners (in their infinite wisdom) have introduced AI into nuclear decision-making to “rectify” that.

Presumably, there would still be human decision-makers in this loop. But, aided by AI, human analysis would run in parallel with AI’s more rapid analysis thereby speeding up the entire process and greatly increasing chances for a tragic error. There is also the possibility (or likelihood) that this use of AI would circumvent the programming that prevents AI from actually making the launch order itself.

The question to consider here is: Will each of these AI systems be able to autonomously engage in secret planning and discussion with the other AI systems? And will problem solving be interpreted as the need to take their own initiatives? The Army and other branches will certainly be able to pass information between AI systems. However, given the rogue hacking behavior we’ve already seen, it seems highly likely that these systems will be able to establish their own clandestine communications and interact with each other independent of even sophisticated military oversight, perhaps even overriding established nuclear safeguards.

For now, these interconnected AI systems are designed to do everything except make the final decision to actually launch missiles with nuclear warheads. But even before militarized autonomous AI, the danger of human error existed. Seen from a real-world perspective, while AI is doing its analysis in a few minutes, a handful of military commanders will be doing theirs. They will have no time to review giant data sets. They will have to go with what they already know and feel and be able to very quickly confer remotely with each other. Human decision-makers will see what actions AI recommends, but will probably not have the time needed to review AI’s rationale, except for a few brief summary bullets: a leap of faith, with humanity’s future hanging in the balance.

History can be instructive in this regard. In 1979, Deputy Secretary of Defense Admiral William Perry was awakened by a 3:00 am phone call alerting him that US warning systems showed 200 Soviet ICBMs barreling towards the US. Perry had about nine minutes to decide whether or not to tell the president to launch a counterattack. If he took any longer, Soviet warheads would strike US ICBMs before they could be launched.

Fortunately, Perry kept a cool head realizing that relations between the US and the Soviet Union were, at the time, calm and diplomatically on solid ground. With the fate of the Earth hanging in the balance, he concluded that such an attack didn’t make sense. He made and explained his recommendation—do not launch. Forty-eight hours later it was determined that a computer simulation of the Soviet attack was responsible for the error. This is a powerful object lesson as to why we should not give away human agency to mere machines.

Red Alert: How Likely Is AI Armageddon?

In the narrative outlined previously, Anthropic lost its contract because it insisted that its AI system could not be used by the military for domestic surveillance or autonomous weapons. The Pentagon then contracted with OpenAI. When Anthropic refused to sign onto these terms over their concerns, the Pentagon irrationally designated Anthropic a "supply-chain risk," leading to a legal battle and cancellation of its contract in October 2026. In the meantime, DOW came up with another legal workaround called “lawful use.” This broad, vague language avoids specific reference to domestic surveillance and autonomous weapons systems, and is designed to provide the public with assurances that AI guardrails are in place.

Eight of the world's leading AI firms including SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon, and Oracle agreed to deploy their advanced AI capabilities on DOW classified networks for “lawful operational use.” But astonishingly, as far as the Pentagon and the Trump administration is concerned, the genocide in Gaza is considered “lawful use.” In the aftermath of this debacle, Sen. Elizabeth Warren (D-Mass.) sent the Pentagon a letter demanding that Congress and the public be informed of the nature of the AI work and that it disclose any safety guardrails being implemented. The letter went unanswered.

Perhaps the only hope here is that the massive resistance against AI we’re seeing in many nations and on a bipartisan basis in the US will ultimately create a powerful global change.

The use of tactical nuclear weapons—the size of the Hiroshima bomb or smaller—has recently been integrated into the battlefield war-fighting doctrines of nuclear-capable nations. It’s generally assumed that if nuclear weapons use is introduced into the current Ukraine or Middle East conflicts, it will involve one or more tactical nukes. However, as war games have shown, the use of tactical nukes can quickly escalate to the horrendous devastation of full-scale nuclear war with strategic warheads over 20 times more powerful than the Hiroshima bomb.

If US President Donald Trump, Russian President Vladimir Putin, or Israeli Prime Minister Benjamin Netanyahu were to approve using a tactical weapon, they would do so knowing that the likelihood of Armageddon was a very serious risk. In the convoluted thinking associated with leaders who have little regard for the best and most humane values of Western civilization, using AI as the decision-maker in these situations could give them a reprieve from the damning judgment of history. It would alleviate their responsibility for either WWIII or much wider scale nuclear devastation. Perhaps the only hope here is that the massive resistance against AI we’re seeing in many nations and on a bipartisan basis in the US will ultimately create a powerful global change in awareness acting as a strong counterweight to this latest, more “advanced” version of nuclear madness.