AI Whistleblower: We Are Being Gaslit By The AI Companies! They’re Hiding The Truth About AI!

A lot of what’s happening in AI right now feels deeply inhumane.

Playing devil’s advocate: if the group that moves fastest on AI wins, shouldn’t we race?

That’s a profitable myth, and I’ve seen plans to stoke it so the public cedes power and resources to a few firms.

We should break up AI empires because they seize people’s data and creators’ work, exploit global labor, drive environmental costs, crush scrutiny, and even muzzle inconvenient researchers; the tools can help, but the way they’re built is harming people, and there are better ways to develop them.

Quick favor before we dive in: hit follow so the best episodes surface in your feed.

Karen, your book is Empire of AI; what journey led you here?

I trained in mechanical engineering at MIT, joined a climate tech startup in Silicon Valley, and watched profit trump public good, which pushed me into AI journalism at MIT Technology Review.

Since 2018 I’ve interviewed hundreds of people to ask who decides what gets built, how money and ideology steer it, and how we redesign innovation to serve broad human flourishing.

How big was your reporting effort?

Over 300 interviews with more than 250 people, including around 90 current or former OpenAI staff, and I went far beyond Silicon Valley to see where the industry’s rhetoric breaks against reality.

Let’s keep this simple for newcomers; where should we start?

AI launched as a field in 1956 with a label that tied it to recreating human intelligence, even though there’s no scientific consensus on what human intelligence is.

That vagueness lets companies redefine AGI for whatever audience they need—curing cancer for Congress, a perfect assistant for consumers, a giant revenue engine for investors—which keeps moving the goalposts.

In 2015 Sam Altman warned AI could be the greatest threat; what was going on?

He was courting Elon Musk to co‑found OpenAI and mirrored Musk’s alarmist language to win him over, even pivoting from his prior focus on engineered viruses.

So did he manipulate Musk?

Musk says he felt played, and documents show he was later muscled out; emails also show Ilya Sutskever and Greg Brockman initially favored Musk as CEO of the for‑profit, then switched after Altman warned he’d be an unpredictable risk.

What do you make of Sam Altman?

People are polarized: if you share his vision, he’s a master recruiter and storyteller; if you don’t, he feels manipulative, which is why leaders like Dario Amodei left feeling used.

Why did Ilya leave?

He helped move to fire Altman, believing Altman’s behavior undermined both safety and the push toward AGI by pitting teams against each other and sowing chaos; after Altman returned, Ilya didn’t.

He once compared humans to animals on a highway project; how do beliefs about intelligence shape this?

Ilya and Hinton treat brains as statistical engines, so they believe scaling similar models will surpass us, but that’s a contested hypothesis, not settled science.

Does the mechanism matter if the outcomes are useful?

It matters because their hypothesis justifies data grabs, massive compute, and labor exploitation to chase human‑replacement goals; we could instead aim for tools that improve health or science without copying a brain.

Why pursue human‑replacement at all?

Because these firms operate like empires.

Define that.

They claim resources that aren’t theirs—our data and creators’ IP—use vast, precarious labor to build systems that also automate labor rights away, and monopolize knowledge by bankrolling and shaping most AI research.

So they’re gaslighting the public?

Yes; look at Google firing Timnit Gebru after she co‑authored a paper on language‑model harms, or companies setting research agendas while dismissing critical findings.

And journalists?

OpenAI even subpoenaed a watchdog, seemingly to intimidate and map critics, fishing for nonexistent Musk ties during its nonprofit‑to‑for‑profit pivot.

You also say empires sell a story of being the ‘good’ empire.

They promise AI heaven if we let them rule and warn of hell if China or, earlier, Google gets there first; that fear justifies hoarding power.

Do leaders think it all ends well?

They lean on worst‑case extinction and best‑case abundance to argue only they should drive, which is an anti‑democratic stance dressed as safety.

Altman tweeted about two ‘fair’ books and an unnamed third; was that yours?

Yes; OpenAI first agreed to help with my book, then shut the door after the board drama, and ignored forty pages of questions.

He does many interviews but not mine; why?

Access is a lever—companies dangle appearances to shape coverage and punish platforms that host critics, which is why I learned to report without the front door and still did more than 300 interviews.

Why did the board fire him and then reverse?

Ilya confided concerns to independent director Helen Toner, Mira Murati added evidence, and independents weighed the stakes of chaos after ChatGPT’s viral launch.

They also found OpenAI’s startup fund paperwork made it Altman’s fund, not the company’s, which deepened trust issues; fearing his persuasion, they moved fast, blindsided stakeholders, sparked backlash, and then reinstalled him.

It’s chilling when founders say someone shouldn’t have their finger on the button.

Both Ilya and Mira said he wasn’t the right leader for that responsibility, and neither returned; many early hires left after clashing with him.

Ilya then launched Safe Superintelligence.

It’s no accident every tech billionaire spins up an AI lab—Musk with xAI, Dario with Anthropic, Ilya with SSI, Mira with Thinking Machines—each wants AI in their own image.

Are they ‘summoning the demon’ for power and legacy?

The demon talk is myth‑making to centralize control—‘if we don’t build it, doom; if we do, utopia’—and it works.

Like in Dune, they seed myths, then start believing them; that blur, plus constant fundraising, creates cognitive dissonance where scary odds and rosy promises coexist.

Is any CEO clearly more moral?

Swapping leaders won’t fix a system that concentrates world‑shaping choices in a few hands while those affected lack a say, and these firms spend heavily to mold the laws that govern them.

Back to the race with China: doesn’t more intelligence mean national supremacy?

That assumes scaling equals general intelligence, but today’s models are jagged—good at some tasks, bad at others—and capabilities advance where firms invest data and human training.

Hinton’s ‘brain as a statistical engine’ is a hypothesis, not consensus; inside companies, capability roadmaps follow revenue, not a baby‑like march to general smarts.

Self‑driving shows shared learning at scale.

Those systems still retrain per city and share the same failure modes when they learn the wrong thing, whereas human diversity adds resilience.

Output is what matters; if a car drives safer than a human, who cares how it thinks?

You just made the industry’s favorite prediction—universal gains with scale—that conveniently enriches those selling it.

Elon’s building a supercomputer to brute‑force more generality.

They’re brute‑forcing models they can sell to automate lucrative tasks, not unlocking a coherent, generalized mind.

He says surgeons will be obsolete soon; true?

Bold timelines keep missing—remember the ‘no more radiologists’ claim—and the best results today come from human experts using AI as a tool.

Will most cars be autonomous in the near future?

No; these are probabilistic systems that make irreducible errors, face domain‑shift limits, and must also clear social and legal hurdles about trust and liability.

Are mass layoffs coming?

We’re already seeing big effects driven by both model capabilities and executive choices to replace ‘good enough’ work, sometimes prematurely, as in the much‑debated Klarna example.

At Klarna we let AI handle a large share of support, reduced headcount mainly through attrition, doubled revenue, and saw coding shift dramatically toward machine‑assisted work; I’m optimistic long term, even if the near term is bumpy.

Thoughts on all this?

Leaders can choose to spend more time in person, but many displaced workers are funneled into lower‑paid data annotation, now one of the fastest‑growing roles, to train the very systems that erased their old jobs.

For listeners, data annotation is the tedious labeling and feedback that teaches these models how to respond.

Reinforcement learning leans on armies of data annotators, and the way that work is organized strips people of time, pay, and dignity. I shared a mother’s story of racing to click tasks while yelling at her child, then realizing she felt turned into a machine so others could enjoy more human jobs.

If disruption hits knowledge fields fast, the internet lets AI scale instantly, leaving people little time to retrain.

That speed is a choice made by companies locked in a race, and it steamrolls the people in the way.

So what happens to them—purpose, pride, mental health—when work vanishes, like the doctor who became a cleaner and spiraled; scaled up, that pain spreads.

This is why I argue AI is widening the gap: the haves gain time while the have-nots get squeezed, including through environmental harm. Think mega data centers sited in vulnerable places—an Abilene project the size of Central Park with roughly a million chips and a power draw beyond twenty percent of New York City.

I thought earlier you were skeptical about the job-loss forecasts.

I’m skeptical of executive hype as a tactic to shape public consent, but the economy is already restructuring.

These facilities strain grids and drink fresh water that towns also need; in Memphis, a supercomputer ran on dozens of methane turbines that neighbors discovered by the fumes, worsening asthma and cancer risks in a community already burdened by pollution.

That’s the split: worse jobs, higher bills, dirtier air for many, while others prosper—so what do we build instead? Use the transportation lens and favor bicycles of AI—small, targeted systems like AlphaFold—that deliver big public value with far less compute, energy, and emissions.

But hasn’t the horse bolted on data and IP, with platforms now offering one click model training on creators’ work?

If the horse were gone, they would not still be devouring data; the appetite keeps growing, and demand for human labeling has risen over years.

Will brute-force scaling slow or change course?

Rather than waiting for that, ask what we should do now, because these firms still need labor, data, power, and public buy-in.

Saying stop feels impossible when governments are cheering them on.

Break up the empire logic and build alternatives; most Americans now want real rules, protests are blocking and banning data centers, artists are suing over IP, and grieving parents have forced a reckoning over child safety after chatbot-fueled abuse.

So what can listeners actually do?

Withhold data where you can, organize against local data centers, shape your school or company’s AI policies, and refuse to make adoption effortless; then help create community-led, lower-impact AI that serves people first.

I live with the split—I love what AI lets me build and still worry about the fallout; can both be true?

Absolutely; keep the benefits and redesign incentives and governance so they do not come with hidden costs.

We need a bigger public conversation outside industry.

It is already happening—in city after city I see packed rooms and local governments tackling these issues head-on.

Final question: what advice would you give a friend with a terminal diagnosis, and how would that differ from what you’d do yourself?

I’d tell them to live for themselves and slow down, even though I am not slowing down.

I am glad you are not; this kind of long, open conversation matters, and your work is helping people act together—thank you.