The AI Risk We Can't Ignore

I build with AI every day. It writes code with me, drafts the first pass of things I later rewrite, sorts through more data than I could read in a lifetime, and runs quietly inside products across a dozen small companies I've started or helped start. I am not a doomer. I am also not a hype-man. I'm a builder, twenty years into moving information around — telecom networks first, then AI platforms — and I've reached the point where I think the honest thing to say is: the risk is real, and pretending otherwise is a choice I'm not willing to make.

This isn't a scare piece. It's the view from inside the tent, written by someone who benefits from the technology and still takes it seriously.

Why 2026 is the year I stopped hand-waving

A few things landed in the same window and changed how I think.

In September 2026, Google Research and HHMI Janelia published the complete wiring map of a male fruit fly's central nervous system in Cell — about 166,000 neurons and 125 million connections. It's the largest brain map by neuron count we've ever made. I was so taken by it that I built a little thing to watch a signal move through one, and later realized it looked exactly like the telecom network maps I stared at for two decades. We can now read biological intelligence at the level of individual wires.

At the same time, AI stopped merely describing the world and started generating it. Models now produce whole playable, explorable environments from a text prompt — "world models." We can write synthetic worlds while we're learning to read real brains.

And then the people building the frontier started saying the quiet part louder. In a September 2026 opinion column for The Guardian, former Google DeepMind researcher Alex Turner argued the field "runs an extremely dangerous race towards superintelligent AI," and put his own personal estimate of an AI takeover at roughly one in three — his words, his guess, and worth reading as exactly that. Months earlier, DeepMind CEO Demis Hassabis had already called for research into AI's threats to be done "urgently" alongside "smart regulation" of "the real risks."

None of this is fringe. Back in 2023, hundreds of researchers — including Sam Altman, Dario Amodei, Geoffrey Hinton, and Yoshua Bengio — signed a one-sentence statement that mitigating "the risk of extinction from AI" should be "a global priority alongside other societal-scale risks such as pandemics and nuclear war." The man often called the godfather of AI, Hinton, told 60 Minutes he couldn't "see a path that guarantees safety." Altman himself wrote, back in 2015, that superhuman machine intelligence is probably "the greatest threat to the continued existence of humanity" — and then went and built one of the companies racing toward it.

When the people with the most to gain from a technology are also the ones warning about it, I don't reach for panic. I reach for attention.

What actually worries me

Here's where I try to be precise, because vague dread is useless to a builder.

The near-term harms are already here, and they're mundane. A deepfake video call impersonating a company's CFO helped scammers steal $25.6 million from the engineering firm Arup. Deepfake incidents jumped tenfold worldwide between 2022 and 2023, and Deloitte projects US generative-AI-enabled fraud losses could reach $40 billion by 2027. This isn't science fiction about rogue superintelligence. It's my mother getting a phone call in my cloned voice.

The effect on kids keeps me up. Common Sense Media found nearly three in four US teens have used an AI companion, and the US FTC has opened a formal inquiry into companion chatbots after litigation tied one to a teenager's death. Canada, where I live, still has no comprehensive federal AI law after Bill C-27 died on prorogation. We are running a live experiment on children with no referee.

The economic churn is real, even if the headlines oversell it. The World Economic Forum projects 92 million jobs displaced and 170 million created by 2030 — a net gain, but with 39% of core skills transformed. The IMF estimates about 40% of global jobs are exposed to AI. "Exposed" is not "destroyed," and I refuse to say 300 million jobs will vanish — that's not what Goldman Sachs' oft-misquoted number means. But the transition is going to hurt real people I know, including artists and writers.

And the cost is physical. Data centres already draw about 1.5% of the world's electricity, on track to roughly double by 2030. The water numbers are more moderate than the scary version — Google now estimates a median text prompt uses about a quarter of a millilitre, five drops — but at planetary scale, small numbers add up.

What worries me less

I try to keep my fear calibrated, because miscalibrated fear is just noise.

I don't lose sleep over the viral stunts — the "fly brain playing Doom," the clips of a connectome "trading crypto." They're entertaining internet culture, not evidence of a digital soul. The verifiable science is the connectome dataset itself, not the meme. I also ignore the recycled line that our attention spans have dropped "below a goldfish's eight seconds" — that's a debunked marketing myth. And I treat the more dramatic claims circulating online — that AI will quietly shave years off human life expectancy, or that some unreleased mega-model already exists — as unverified until someone shows me a real source. There isn't one.

The point of taking risk seriously is that you have to be equally serious about not inflating it. Credibility is the only currency that matters in this conversation.

How I choose to build

So what do I actually do, running ventures that all touch AI? A few rules I hold myself to:

Be honest about what's a prototype. When I built that fly-brain visualization, I said plainly it was a stylized model, not a living brain. The instinct to oversell is the original sin of this field. I'd rather under-claim and be trusted.

Keep a human in the loop where it counts. AI drafts; people decide. For anything that touches a customer's money, a child, or someone's livelihood, the model is an assistant, never the final authority.

Build for observability and orchestration. My telecom years taught me you can't govern what you can't see. The same lesson from watching a brain route signals applies to AI systems: instrument everything, know when a model is drifting, and be able to shut a thing off. That's not bureaucracy — it's basic respect for the people downstream.

Pace yourself on purpose. When the labs building the frontier are asking for more caution, the right move for the rest of us isn't to sprint faster into the gap. Ship slower than you can. Test more than feels necessary.

I'm still an optimist about tools — I build them for a living, and I've watched AI genuinely help people. But the exciting version and the dangerous version of a technology are usually the same version, just pointed differently. The honest response isn't hype and it isn't doom. It's paying attention early, while a fruit fly is still the headline and not the afterthought.

TheAIRiskWeCan'tIgnore

Why 2026 is the year I stopped hand-waving

What actually worries me

What worries me less

How I choose to build

More from across our network