Musk Wiki

The Shifting Remedy

NextThe Why Behind Mars

The Shifting Remedy

For more than a decade Elon Musk has said the same thing about advanced AI: it could end us. What changes is his answer. The fear holds still while the proposed fix cycles through four shapes. First, warn the field and slow it down (2014–2018). Then ask a government to referee it (2015–2023). Then build a safer AI yourself (2023). Finally, give up on control altogether and bet on the values you instill plus the pressure of competition (2024–2025).

Two long-running puzzles resolve along the way. One is the apparent contradiction between Musk demanding a new AI regulator and Musk crusading to gut regulators in the same years. The other is a stale read of his position: take “he wants a regulator” as his standing 2026 view and you’d be wrong, because by 2024–2025 he had moved past it.

Summary

The fear is constant; the fix is not. From 2014 Musk treats advanced AI as a civilization-level danger. His first answer is to warn the world and pump the brakes, a campaign he later calls a decade spent as Cassandra, and that by 2018 he concedes had failed (“I tried for years”). His second answer runs in parallel from 2015 to 2023: hand the job to an outside referee, a government agency that watches the labs the way aviation and auto regulators watch theirs. In 2023 he keeps pressing for that regulator but adds a third move on top of it — build the safer AI himself. That lab is TruthGPT, soon renamed xAI, a truth-seeking shop meant to do, from the inside, the job no regulator ever took up.

Then comes the largest turn. By 2024–2025 he drops the oversight ask entirely, because he no longer thinks control is even possible. A chimp can’t control humans; “the AI is going to be in charge … not humans”. If you can’t steer it, the only levers left are the values you build into it and the competition that one honest player forces on everyone else. So the remedy migrates from oversight to participation to letting go: warn it, regulate it, out-build it, then stop trying to drive and just try to make it friendly. The worry that the wrong AI wins never moves. What moves is that each fix, by his own account, ran out, and the next one took its place.

2014–2018 — warn and slow it down

The first move is to hit the brakes. The danger enters his public voice in 2014, with the “more dangerous than nukes” tweets. His earliest dated prescription comes in the January 2015 Reddit AMA, eleven months before he co-founds OpenAI, and it already sets the two-part shape the whole arc keeps. The danger is real but not around the corner, and the response is concrete safety work:

“The timeframe is not immediate, but we should be concerned. There needs to be a lot more work on AI safety.” ↗

By the 2018 Joe Rogan conversation he is at his bleakest, because by his own account the braking had failed. He had spent years trying to get the field to slow down, and nobody slowed:

“I tried to convince people to slow down, slow down AI, to regulate AI.” ↗

“I tried for years.” ↗

That is the low point that closes stage one. “Slow down” never took. The fuller second remedy, an outside referee, is what he carries into the years ahead.

2015–2023 — ask for an external referee

Alongside the warnings, from 2015 on, comes the remedy Musk asks for most often: a third party to watch the labs, because the people building the thing can’t be the only ones checking it. In 2016 he gives the structural version, spread the technology so no single actor controls it. The institutional version is a government regulator, and he repeats it year after year in almost the same words.

At the 2017 World Government Summit his first piece of advice to governments is exactly this:

“governments keep a close eye on artificial intelligence and make sure that it does not represent a danger to the public.” ↗

In the 2019 Lex Fridman conversation (#49) he sharpens the ask into a named agency:

“Where there is a lack of investment is in AI safety, and there should be, in my view, a government agency that oversees anything related to AI to confirm that it is, does not represent a public safety risk.” ↗

At TED2022 he puts it most plainly, and notably he is asking to be policed against his own interest:

“I do think there should be a regulatory agency for AI.” ↗

“I don’t love being regulated, but I think this is an important thing for public safety.” ↗

Five months later, at Tesla AI Day 2022, he aims the same demand straight at AGI and calls for a referee:

“I think there should be a referee that is trying to ensure public safety for AGI” ↗

By 2023 the rest of his politics has turned hard against regulation, yet on this one subject he keeps correcting the caricature. He is not the anti-regulation maverick people assume. From the April 2023 Tucker Carlson interview:

“Some people may think I’m some revelatory maverick that defies regulators on a regular basis. This is not the case.” ↗

In the same interview he offers his proof that he means it. The one private meeting he ever got with a sitting president, he spent on AI regulation:

“I saw it happening from well before GPT-1, which is why I tried to warn the public for years. The only one on one meeting I ever had with Obama as President I used not to promote Tesla or SpaceX, but to encourage AI regulation.” ↗

The Bill Maher interview that same month puts the oversight ask in its plainest 2023 form:

“there should be some regulatory body that oversees what companies are doing so that they don’t cut corners.” ↗

The case has its sharpest form in an analogy he reaches for. We already regulate nuclear bombs, he points out, and he ranks AI as more dangerous than those. So the precedent for an AI referee is one society has already set.

For eight straight years, 2015 through 2023, the outside referee is the steady center of the arc.

⚠️ The deregulation paradox (resolves C3). This same 2023 window is exactly when Musk the deregulation crusader is loudest. He is firing off the slogans. “There’s no garbage collection for rules,” he says. The country is “Gulliver tied down by a million little strings.” Cut the dead weight, “if we get rid of nonsense regulations … a golden age.” His sharpest version aims at space: “There is simply no way that humanity can become a spacefaring civilization without major regulatory reform.” Asking for a new federal AI regulator while campaigning to scrap regulators wholesale looks like a flat contradiction. His own positions reconcile it through a distinction between domains. AI is a public-safety field like aircraft, cars, food and drugs, the analogy he leans on for years, so it warrants a referee. General bureaucratic sprawl is just rule-mass to be cut back. He never lines the two up in a single source, but the logic is implicit in how he argues each side. They reconcile on that line. What really dissolves the tension, though, is that the AI-regulator ask itself didn’t survive: by 2024–2025 he stops asking for oversight at all (next two sections). The deregulation stance is left standing; the AI exception is what gets dropped.

2023 — start building (while still asking)

By 2023 the referee still hasn’t shown up, so Musk doesn’t swap one fix for another, he stacks a new one on top. He keeps pressing for regulation, as the quotes above show, and at the same time sets out to build the safer lab himself. He names the new move in that very same April 2023 Tucker Carlson interview, almost in the same breath as the regulation ask. The constructive answer is a product, “TruthGPT”:

“I’m going to start something which I call TruthGPT, or a maximum truth-seeking AI that tries to understand the nature of the universe.” ↗

Three months later TruthGPT becomes xAI, and he states its whole safety case as a design goal in the 2024 Lex Fridman conversation:

“I think it’s important that whatever AI wins, it’s a maximum truth seeking AI that is not forced to lie for political correctness, or, well, for any reason, really, political, anything.” ↗

The move rhymes with an earlier one. He had founded OpenAI to be the antidote to an unsafe lab (the OpenAI arc); now he runs the same play one step further out and builds xAI. The weight is shifting from oversight, a regulator watching the builders, toward participation, become a builder and be the safe one. In 2023 the two still travel together. Oversight isn’t gone yet, just joined.

2024–2025 — give up on control; instilled values and competition

The last shift is the biggest. Musk decides that control of a superintelligence is impossible in principle, and that quietly kills the whole oversight idea: you can’t referee what nobody can control. In the October 2025 Joe Rogan conversation he reaches for a chimp:

“I mean, I don’t think anyone’s ultimately going to have control over digital superintelligence, any more than, say, a chimp would have control over humans. Chimps don’t have control over humans, there’s nothing they could do.” ↗

At the November 2025 Tesla shareholder meeting he says it as plain fact, and when someone asks who should hold power, he treats the question as moot:

“Well, I mean, I think actually long term, uh, the AI is going to be in charge to be totally frank, not humans.” ↗

“if if if artificial intelligence vastly exceeds the sum of human intelligence, it is difficult to imagine that that any humans will actually be in charge.” ↗

If no one can be in charge, the fix can’t be control. So the entire regulator program collapses, in that same meeting, into a four-word instruction:

“we just need to make sure the AI is friendly.” ↗

Two things step in where control used to be. The first is the values you build in, above all truthfulness, which he names the single most important safety property in #438:

“the thing that at least my biological neural net comes up with as being the most important thing is adherence to truth, whether that truth is politically correct or not.” ↗

The second is competition. Not that his AI has to win, but that a single honest entrant shames the rest of the field into doing better, the market-pressure theory of safety he lays out on #2404:

“Yes, as long as there’s at least one AI that is maximally truth-seeking, curious, and for example, weighs all human lives equally, does not favor one race or gender, and people are able to look at Grok and xAI and compare that and say, wait a second, why are all these other AIs being basically sexist and racist? And then that causes some embarrassment for the other AIs and then they improve.” ↗

This is where the arc lands. Not “regulate it,” not even “out-build it,” but build in the right value, truth, and let competition police the rest. Steering it has given way to seeding it and racing it.

⚠️ The remedy moved; the fear held (supersedes ST2). It’s tempting to read “he wants an external referee” as Musk’s current ask. It isn’t. The regulator was the 2015–2023 remedy. By 2024–2025 instilled-values-plus-competition has taken its place, because he no longer thinks control, and so oversight, is possible at all. The danger he names holds steady across the whole decade. Only the proposed fix gets superseded, one stage at a time.

What the arc reveals

  • The fear is fixed; the remedy is serial. Across eleven years, from the 2014 “more dangerous than nukes” to the 2025 “AI is going to be in charge”, the threat he describes barely moves. The fix runs through four incompatible shapes: slow it, regulate it, out-build it, stop controlling it. His other AI story, the OpenAI arc, is the vehicle reversing while the conviction held. This one is the conviction holding while the cure kept failing and getting swapped out.
  • Each remedy is dropped because the last one ran out, not because the worry eased. “Slow down” goes because it failed (“I tried for years”). The regulator goes because no one built it, and because he decides control is impossible anyway. Building the safe lab narrows to instilling a single value, because even the builder can’t stay in charge. It reads as a steady retreat in the face of one problem he can’t crack.
  • It settles the corpus’s one cross-cluster contradiction. The AI-regulator ask here (2015–2023) and the deregulation crusade of the same years only clash if you read both as current. The domain distinction reconciles them in principle. Dropping the AI-regulator ask resolves them in fact, leaving the deregulation stance intact and the AI exception retired.

Connections