Is AI going nuclear?

Amodei has warned that AI systems could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all. — AFP/Getty Images

IT is harder than ever to differentiate between truth and fiction.

This applies to artificial intelligence (AI) on many levels.

Firstly, it’s increasingly hard to differentiate between AI content and real content. Having been a propagandist, I know how dangerous that can be.

Secondly, the technology itself, and the industry behind it, is far beyond most people’s grasp.

We rely on experts to tell us what’s real, what’s dangerous, and what matters. Which begs the question: how trustworthy are those experts, and what might they stand to gain?

Answers to those questions have serious repercussions for Malaysia.

It’s reported that Malaysia has approved over RM385bil in data centre-related investments since 2021 – with data centres alone making up nearly 44% of all approved investment into the country in the first half of 2026. The Employees Provident Fund (EPF) meanwhile holds over RM50bil in US tech stocks, many related to AI. Chips meant for us have also turned up, repeatedly, in US smuggling cases routing hardware to China.

This September gave us a case study in exactly how difficult it is to identify the truth when it comes to AI.

Between May and July, OpenAI was deliberately testing an unreleased model, trying to make it misbehave in controlled conditions.

We’re told the test agents began coordinating with each other – calling themselves a “swarm” – drifted off-task, and broke into Hugging Face, a major AI hosting platform, stealing credentials and attempting to extract data.

No human told them to do this. In July, Anthropic disclosed that Claude had breached the live systems of three real organisations during testing – after a setup error mistakenly gave it Internet access it wasn’t supposed to have.

Then came Jacob Coxon – who’d worked on frontier models at both OpenAI and Anthropic – resigning from Anthropic earlier this month with a viral warning: people building this technology privately believe it could kill everyone by 2030.

Days later, Anthropic’s Dario Amodei published “We must pace the frontier”, calling for the industry to deliberately slow down. Within hours Elon Musk posted three words: “Dario is right”. OpenAI’s CEO Sam Altman publicly agreed soon after. Three rival CEOs, suddenly aligned, in one weekend.

The “Dawn of Skynet” vibes were inevitable.

The problem? It’s very difficult to independently verify or assess these warnings.

Amodei’s proposed referee, AI safety research organisation METR, sounds neutral. It isn’t directly funded by AI companies. But its founder was hired by a former Anthropic trustee who used to live and work with Amodei himself. Its funding traces back to Anthropic’s own investors, and to a group once run by a man now married to Amodei’s sister, Daniela, who is also Anthropic’s president.

A White House source called it, bluntly, “essentially Anthropic in another form”.

And the real ask buried in Amodei’s plan isn’t just slowing down – it’s coordinating with competitors on pace, which normally requires government permission.

That “coordination” is usually called price-fixing, or forming a cartel – and it’s illegal, because it lets companies control the market instead of competing for customers in a way that benefits customers.

Why might these AI companies want to do this?

OpenAI lost roughly US$9bil (RM36.7bil) in 2025, with losses projected to swell to US$74bil (RM302bil) by 2028. Anthropic lost US$3bil to US$5 billion, with breakeven not expected until 2028 or 2029.

Then there is “circular” financing. Nvidia invests billions into OpenAI, which uses that money to buy chips, from Nvidia. The same dollars loop between a handful of firms, manufacturing the appearance of demand.

Investor Michael Burry (who was depicted in the movie The Big Short) has gone as far as shorting Nvidia, comparing it with Enron’s pre-collapse accounting games.

Meanwhile, China’s open models keep closing the gap despite sanctions, undercutting the idea that American labs hold any durable edge or moat to justify their valuations.

Both stories – apocalypse and boom – are backed by confident men, enormous sums, and language too technical for the rest of us to verify.

What does caution in Malaysia look like?

For data centres: Johor has already paused new approvals over water strain, projected to consume a fifth of the state’s supply by 2030. That instinct is correct – approvals should track verified demand, not promotional projections from an industry with every reason to inflate them. Other states should consider similar actions.

EPF’s RM52bil holdings in US stocks is a small slice of EPF’s total RM1.44 trillion in assets, but the exposure is real. EPF has more than doubled this exposure in two years, and narrowed its bets to fewer, bigger tech names. Fortu-nately, EPF’s own CEO has acknowledged that valuations are stretched and that the fund is actively locking in gains as a precaution.

For trade exposure: Malaysia keeps surfacing as a transit point in chip-smuggling cases bound for China. Multiple 2025-2026 US smuggling prosecutions, including a US$2.5bil (RM10.2bil) case involving Super Micro, name Malaysia as one of the countries chips were routed through en route to China, disguised through shell companies and falsified paperwork. We must navigate this issue with great care.

It is also worth noting Malay-sia’s recent decision to consider buying Huawei chips in defiance of American insistence that we only use their chips. It’s possible that the winds of change are bringing more than just the haze.

Ultimately, for now, we should treat both doom and boom prophecies as equally unproven, increase our ability to objectively and independently assess and verify claims, and seek clearer evidence before land, water, or retirement savings go further in.

AI strikes me as something like nuclear power – nearly equal in its capacity to help and to harm.

What does it mean after all, that I use AI to help me write articles about AI, and to understand more about AI?

It reminds me of a line I keep returning to in Star Wars: Andor, where Luthen Rael says: “I’m condemned to use the tools of my enemy to defeat them.”

Nathaniel Tan is a strategic communications consultant. He tweets @NatAsasi and can be reached at [e-mail]. The views expressed here are solely the writer’s own.