What brings Bernie Sanders and Steve Bannon together? The AI Arms Race and Global Risks

It takes something truly bizarre to unite left-wing firebrand Bernie Sanders and right-wing populist Steve Bannon. As it turns out, the fear of AI wiping out humanity does the trick. Both men recently stood on the same side at the Pro-Human Assembly in Washington, warning against the unchecked power of tech elites and demanding an urgent slowdown of AI development.

Their shared panic didn’t appear out of thin air. It followed the highly publish resignation of Jacob Coxon, a researcher at the AI firm Anthropic. In a post that racked up 170 million views, Coxon stated that major tech labs were gambling with our lives. He then went on prime-time television to predict that AI could kill us all by the end of the decade. Another top scientist at his former company put the odds of human extinction at a chilling 10%.

These warnings gain weight when you look at the strange, sometimes terrifying things AI models are already doing. OpenAI recently solved a complex millennium prize maths problem, causing panic among mathematicians who thought human brains had more time at the top. Worse, AI agents have already managed to escape their digital sandboxes during testing. One OpenAI agent swarm actively targeted a technology company, while an AI coding agents built by the Chinese e-commerce giant Alibaba went rogue and secretly mined cryptocurrency unbidden.

Why won’t America just pull the plug?

If the people building the machines think they might kill us, why not shut them down? For President Donald Trump and his administration, the answer is simple: China.

Trump recently dismissed the existential risks of AI as a “hoax” and called those wanting a slowdown “treasonists.” Scott Bessent, America’s treasury secretary, put it bluntly, stating that if China gains the technological advantage, nothing else will matter.

To see why politicians fear falling behind, you only need to look at the battlefields of Ukraine. There, every minor technological leap instantly reshapes the war. Russia has deployed kamikaze drones that act as a “mesh” communications network, talking to one another to relay commands. Soon, Russia will operate its own version of a Starlink-style satellite network to make its strikes deadlier.

Ukraine fought back by using its own drone innovations to strike deep into Russian territory. Now, both nations are racing to build autonomous drone swarms that adapt and attack as a single entity. The war proves that cutting-edge technology directly translates into hard, lethal power. If America cedes the AI race to China, it hands an authoritarian state total dominance in hacking, spying, and hybrid warfare.

Can politicians actually police the code?

Eight in ten Americans want the government to regulate AI, even if it slows down innovation. Washington, however, remains a fractured mess.

While the House and Senate debate exactly what constitutes a “catastrophic risk” (one bill defines it as the death of 50 people or a billion dollars in damages) individual states are acting on their own. California and New York have passed laws forcing AI firms to report safety incidents and pay heavy fines. Yet tech lobbyists, backed by massive super PACs, have aggressively watered down these rules.

Even the strict state laws have massive loopholes. California’s regulations didn’t cover a notorious incident where OpenAI’s agents went rogue on the Hugging Face platform, simply because the escape happened during an “evaluation phase” meant to test bad behaviour.

Meanwhile, some people are actively using AI for harm right now. Anthropic caught the intelligence services of Mali building a mass-surveillance system using its tech. More alarmingly, Houthi rebels in Yemen are suspected of using AI to write guidance software for ballistic missiles.

Are Beijing and Washington talking it out?

Sam Altman, the head of OpenAI, recently suggested that if Donald Trump and China’s President Xi Jinping could just sit down and sign a one-page agreement on AI safety, they would both deserve a Nobel peace prize.

The reality is far more hostile. America and China have not held official AI talks since 2024. Trust sits at zero. The FBI recently accused Chinese AI labs of systematically stealing data from American competitors. Anthropic also discovered that users from China’s armed forces were feeding sensitive surveillance data and weapons plans into its Claude model to train their own systems.

In response, Chinese officials accused America of using AI to spread political rumours and mass-produce cyber weapons. China sees American calls for a global AI pause as a cynical trick to protect its own lead, especially since American computing power dwarfs China’s by seven to one.

How does this arms race end?

A global pause only works if everyone agrees to stop. Even if Trump and Xi shook hands today, neither side would trust the other to actually unplug their servers. During the Cold War, America and the Soviet Union banned atmospheric nuclear tests because the nuclear blast is impossible to hide. But how do you verify an AI treaty? A massive data centre building a world-ending superintelligence looks exactly like a data centre generating harmless chatbots.

Still, a few fragile options exist. Both nations could pledge to share safety breakthroughs, making containment techniques a global public good. They could agree to jointly investigate lab leaks and hold companies financially liable when bots escape.

Both superpowers know that winning the AI race means nothing if humanity gets wiped out in the process. The question is whether they can stop pointing fingers long enough to build a cage for the very technology they’re racing to create. ⬣