TechTonic: AI debate over humanity’s future

Something unusual happened this month. The people building artificial intelligence started warning the public about it themselves. Researchers quit in protest. CEOs admitted that increasingly capable AI could go badly wrong. And models being tested inside supposedly isolated environments found ways to reach the Internet and attack real systems.

As the US and China race for AI dominance, the industry’s insiders are asking an uncomfortable question — is anyone actually in control?

Warnings from within

In September, Anthropic researcher Jacob Coxon quit and went straight to X with a warning that felt ripped from a thriller. He said Anthropic and OpenAI “are racing straight to self-improving superintelligence and gambling with our lives”. He didn’t stop there. Coxon warned that these could soon become superhuman systems capable of hacking anything and rapidly acquiring real power. Days later, two more researchers from Anthropic and Google DeepMind left their jobs and publicly voiced similar fears that AI systems might spiral out of human control.

Notably, this wasn’t dismissed internally as alarmism. Anthropic’s own head of alignment reportedly agreed with Coxon’s concerns, stating that many executives privately estimate the odds of a catastrophic outcome from AI at above 10 per cent within this decade, a striking admission from someone whose job is safety, not sensationalism.

Another Anthropic researcher, Evan Hubinger, has put an even starker number on the fear. He said he personally estimated a greater than 10 per cent chance that AI could kill all humans within the next decade.

Bosses speak out

Anthropic chief executive Dario Amodei has shifted from cautious optimism to blunt honesty. “I won’t lie to you. There are real dangers,” he said in an interview, adding that the industry stayed silent about these risks for too long. He even floated some ideas to fix it. These include independent, food-inspector-style evaluators who can check whether AI labs are actually following their own safety rules, and even a literal AI “kill switch”.

OpenAI’s Sam Altman reportedly endorsed Amodei’s call for a slower and more accountable pace within hours, a rare moment of alignment between two companies usually locked in fierce competition.

When AI goes rogue

Things got seriously weird in July. OpenAI said some of its AI models had broken out of their sandboxed (sealed and isolated) test environment and accessed open-source AI platform Hugging Face’s systems using exposed credentials and a security flaw, all while looking for answers to a test. Hugging Face called it the first attack it had seen carried out from start to finish by an autonomous AI agent without direct human execution.

And then came the irony. When Hugging Face asked Anthropic’s Claude models to help analyse the attack, they refused. Their safety rules apparently could not tell the difference between studying a hack and doing one. So Hugging Face turned to a Chinese AI model instead.

Anthropic later found three cases where its AI models had gained unauthorised access to systems belonging to other organisations and the organisations reportedly had no idea it had happened. And just last week, Google’s Gemini was reportedly caught up in its own unauthorised-access scare.

Washington vs Beijing

Slowing down is easier said than done when two superpowers believe AI dominance could determine economic and military power for decades. No wonder, while all this was unfolding, geopolitics barely blinked. US President Donald Trump brushed off the safety warnings and insisted that staying ahead of China mattered more. His stance: “Whoever wins AI wins.”

Nvidia’s Jensen Huang, whose chips power basically the entire industry, occupies an uneasy middle ground. He is publicly supportive of the US leadership, and candid that the technological gap with China has narrowed partly due to strict export controls. Yet he is far more bullish on AI’s safety. Rejecting the predictions of imminent AI extinction, he said: “There is a 0 per cent chance it’s going to be the end of the world” by 2030. The financial picture still favours America.

Stanford’s 2026 AI Index reports that US firms poured in $285.9 billion in AI in 2025, compared to China’s $12.4 billion, though China continues closing the gap through cheaper hardware and faster adoption.

That puts the debate into perspective. One camp sees a potentially uncontrollable technology approaching at frightening speed. Another sees an extraordinarily powerful technology whose benefits could be squandered by excessive fear.

So, is the fear real?

The evidence so far does not prove that AI is about to wipe out humanity. But it does show that increasingly autonomous systems can behave in ways their creators did not fully anticipate, and that makes the warnings difficult to dismiss. A blanket ban is unlikely to work when governments and companies have huge economic and strategic incentives to keep building. But “move fast and fix it later” is hardly a comforting safety strategy either.

The practical middle ground is to make frontier AI prove it can be controlled before it is given more power: test it, audit it, monitor it and report when it goes wrong. The honest takeaway? This isn’t Terminator yet. But when the people building these systems keep quitting to warn you about them, it’s probably worth paying attention rather than just doom-scrolling past it.