YT YouTube2Text

YouTube2Text → Transcripts → La Maison-Blanche cache quelque chose dont personne ne parle

La Maison-Blanche cache quelque chose dont personne ne parle — Transcript

by Grand Angle · 3,791 words · 651 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00On October 1st, Time magazine published
0:02a long investigative report on Donald
0:04Trump.
0:05Buried in one of the paragraphs, an
0:07official present at the time recounts a
0:08scene that had remained secret until
0:10then.
0:11It took place 10 months earlier during a
0:13private meeting. On one side of the
0:15table, the President of the United
0:16States; on the other, Elon Musk; and
0:18between them, Grok.
0:21For hours, the President of the United
0:24States questioned Musk's AI about his
0:26presidency, his legacy, and how
0:28Venezuelans would react if the U.S.
0:30captured their president.
0:32The machine replied that Nicolás Maduro
0:35was deeply unpopular and that many would
0:37celebrate his downfall.
0:41On January 3, 2026, U.S. forces captured
0:44Maduro at his residence in Caracas.
0:46Scenes of jubilation erupted in the
0:48streets, especially among the diaspora.
0:51And according to Time, Trump came away
0:53convinced of one thing: Grok is
0:54brilliant.
0:58Obviously, there is no proof that Grok
1:00made the decision. Heads of state have,
1:02after all, always listened to private
1:04advisors. What changes here is the very
1:06nature of the advisor. We are not
1:08talking about a diplomat or a minister,
1:10but a commercial product with its owner
1:12sitting in the room.
1:15Meanwhile, Palantir is providing NATO
1:16with a command system that integrates AI
1:18, capable of cross-referencing data and
1:20supporting military operations.
1:25For its part, OpenAI is directly
1:26offering governments AI infrastructure
1:28for their public services, healthcare,
1:31and education, with the ambition of
1:32providing a ChatGPT tailored to the
1:34citizens of each country.
1:38At least 11 countries have already
1:39joined the initiative.
1:41The company is even putting forward a
1:43redistribution proposal, a public fund
1:45whose returns could be paid directly to
1:48citizens to give them a share of the
1:50growth driven by AI.
1:53Also, remember that its boss, Sam Altman
1:55, has already tackled identity with the
1:57World project, which scans your iris to
1:59provide you with digital proof that you
2:01are indeed a human.
2:03Nearly 18 million registered; we are
2:05certainly far from the full human
2:07population, but the intention is clear.
2:09And to regulate this revolution?
2:12Well, on September 29, six giants of the
2:14earth signed a 300-word text at the
2:16White House in which they pledge to
2:19police themselves.
2:21Add it all up, and you have almost all
2:22the organs of a state. All that's
2:24missing is the logo on the passport.
2:27And inevitably, as these companies
2:29accumulate the means to inform us, treat
2:31us, defend us, and assist those who
2:33govern us, one wonders if they might end
2:36up replacing states.
2:39Replacing a state, however, requires
2:41much more than possessing exceptional
2:43technology.
2:44You have to enforce laws, arbitrate
2:46conflicts, fund services, and answer for
2:49an entire population, even when it isn't
2:51profitable.
2:53But what if there were another path?
2:56A world where elected officials would
2:58continue to pass laws?
2:59But to enforce them, they would have to
3:01negotiate with companies that have
3:02become irreplaceable. They wouldn't need
3:04to officially take power. It would be
3:06enough for them to control the means to
3:08exercise it.
3:09A state keeps the title of sovereignty
3:11but leases out the exercise of it.
3:18Obviously, big companies didn't wait for
3:20ChatGPT to have power. They lobby,
3:22control essential infrastructure, and
3:25can become so important that a
3:27government hesitates to let them fail—
3:29the famous "too big to fail."
3:32But their power doesn't free them from
3:34the state. For a simple reason: a
3:36company can die. A state, well, it can
3:38also die, but much more difficultly and
3:41over a much longer time.
3:42In 2008, Lehman Brothers went bankrupt
3:45and disappeared. In 2012, Greece
3:47defaulted on part of its debt. It is
3:50still here with its territory, its
3:52administration—at least in part—and its
3:54power to levy taxes. A bankrupt company
3:57is finished. A bankrupt state is
3:59generally just a crisis that passes.
4:02And these state powers are what AI
4:04giants need more than ever.
4:07Microsoft signed a 20-year power
4:08purchase agreement to allow for the
4:10restart of a Three Mile Island reactor.
4:13But money and a contract are not enough
4:15. The reactivation requires
4:17authorization from the U.S. nuclear
4:19regulator. Microsoft can buy the
4:20production. Admittedly, it cannot grant
4:23itself the right to produce it.
4:25These companies therefore have an
4:26interest in keeping a state capable of
4:28protecting their investments, enforcing
4:30their contracts, and authorizing their
4:32infrastructure.
4:34But needing the state doesn't prevent
4:35you from making the state dependent on
4:36you.
4:37Take the splitting of property rights.
4:40Parents can transfer the bare ownership
4:42of a house to their children while
4:43retaining the right to use it.
4:45Sovereignty obviously cannot be sliced
4:47up like real estate, but the image makes
4:49our problem clear. The state would
4:51retain legitimacy, the law, taxes, and
4:53the signature. Yet, to understand a
4:56situation, prepare a decision, or
4:57execute it, it could depend more and
4:59more on the capabilities controlled by
5:01its suppliers until reaching the point
5:03of no return.
5:05Part of the power would then shift to
5:07those who determine what it can
5:08effectively do, in what timeframe, and
5:10at what price.
5:12It would keep the right to decide while
5:14losing some of the means to decide
5:16otherwise.
5:17That is what a dismembered sovereignty
5:19could look like. But then why would
5:21leaders accept such a situation?
5:24After all, they are the ones who have to
5:26answer to the voters. Well, precisely,
5:28their need to achieve results can be one
5:30of the drivers of this dependency.
5:36Well then, put yourself in a
5:37government's shoes.
5:38You have to improve public services,
5:40contain spending, ensure security, and
5:42so on.
5:44You have a few years to present a track
5:46record, but building your own
5:47capabilities takes time, while a vendor
5:50offers you a tool that works right now.
5:54In August 2025, OpenAI offered US
5:55federal agents 1 year of access to
5:57ChatGPT Enterprise for 1 dollar per
5:59agency.
6:01The AI doesn't take any power away from
6:02a minister. It even promises them more
6:05means to exercise their own.
6:07A government can therefore become more
6:09effective today by preparing a
6:10dependency that its successors will
6:12inherit tomorrow. And that is the whole
6:14problem with political horizons. The
6:16benefits are reaped during the term. The
6:18cost often arrives afterward and someone
6:20else pays for it. The former Governor of
6:23the Bank of England, Mark Carney, called
6:25this the "tragedy of the horizon"
6:27regarding the climate. This dependency
6:29begins precisely at the moment the
6:31organization transforms around the tool
6:33.
6:34Data is adapted to its operation, teams
6:36learn to work with it, and thus certain
6:38internal skills no longer seem to need
6:40maintenance.
6:42Then comes the renewal. The next
6:44government may disapprove of this
6:45situation.
6:46Nonetheless, it must ensure service
6:48continuity, and it is the one that will
6:50have to explain the consequences of a
6:52break. And that, well, that already
6:54exists.
6:56In December 2025, the British Ministry
6:58of Defence awarded Palantir a contract
7:00worth around 240 million pounds without
7:02competitive bidding.
7:04And for good reason, changing providers
7:06would require rebuilding the
7:08architecture, redoing security
7:10validations, and retraining personnel,
7:12with disruptions to military operations
7:14.
7:14Some would go as far as to speak of a
7:16capture of the Ministry of Defence by
7:18Palantir.
7:19In France, it's essentially the same
7:21story. Since 2016, the DGSI has been
7:23using Palantir. At the time, it was a
7:26transitional solution while a French
7:27tool was being developed. Contract
7:29renewed in 2019, 2022, and then at the
7:31end of 2025.
7:33Last May, the Minister of the Interior
7:35admitted to lawmakers that the French
7:36solution had run into a technical
7:38problem.
7:39Its full deployment is now scheduled for
7:412028, meaning 12 years for a transition
7:44, and even then, we aren't sure we'll
7:46make it.
7:48It's a bit like the story of the frog in
7:50the pot.
7:51Obviously, no government would accept a
7:53company showing up and announcing: "
7:55From now on, we are in charge, and don't
7:57worry, you'll make a great straw man."
8:00But precisely, nobody asks them for that
8:02. Each decision seems reasonable. Each
8:04new contract brings something to the
8:06table.
8:06It is their accumulation that eventually
8:08shifts the balance of power, without
8:10anyone ever officially relinquishing
8:12their sovereignty. That is where the
8:14line is drawn. Having a supplier means
8:16asking them for the means to execute a
8:18decision. Becoming dependent means
8:20having to adapt your decisions to that
8:22supplier's conditions due to a lack of
8:24viable alternatives.
8:26The State retains its powers. It can
8:28challenge, regulate, finance an
8:30alternative solution, whatever you like
8:32.
8:33But all of that takes time, and these
8:35services must continue to function in
8:37the meantime.
8:38The supplier, therefore, does not need
8:40to give an order.
8:41It is enough for them to make refusing
8:43these conditions politically more costly
8:45than accepting them.
8:47To this internal pressure is added
8:49international competition.
8:51In December 2025, Trump even goes so far
8:53as to sign an executive order to attack
8:55certain American laws that regulate AI
8:57and even threatens to strip affected
8:59states of certain federal subsidies.
9:02The justification, you know it by heart
9:04. If American companies are too heavily
9:06regulated, China will win the race. And
9:09that is unacceptable. The government can
9:11thus consider that by protecting its
9:13companies, it is protecting the
9:15country's power.
9:16And if these promises of growth,
9:18security, or modernization rely on their
9:20success, it also becomes politically
9:22invested in their expansion.
9:25Well, from that point on, the company no
9:27longer even needs to pressure the
9:29government to defend its interests.
9:31It begins to do so on its own because it
9:34has made the company's success a
9:35condition of its own. And that is the
9:38whole trap.
9:39The day it wants to impose limits on the
9:41company, it risks weakening the very
9:42means it relies on to govern.
9:45And then there is another reality that
9:47is perhaps even harder to undo: economic
9:50growth. It is estimated that the rise of
9:53AI contributes to about a third of the
9:55year's American growth, taking into
9:57account investments and certain effects
9:59on consumption. And we are only at the
10:01beginning. According to A1, the adoption
10:04of AI in businesses remains largely
10:07superficial. A third of the growth while
10:09this technology is only just beginning
10:11to be integrated. Imagine the political
10:14weight it could carry tomorrow. Here, we
10:16head into an entirely different
10:17implication. The expansion of these same
10:19companies also contributes to the
10:21economic results on which the government
10:23will be judged.
10:24Obviously, this does not mean Washington
10:27will save every struggling AI giant, but
10:29if their slowdown threatens part of the
10:32country's growth, imposing constraints
10:34on them also becomes a political risk.
10:37Interests can therefore converge for a
10:39long time before a conflict reveals the
10:41dependency.
10:46But at this point, you might tell me all
10:48of this already exists. Tech giants have
10:50long held colossal power over states. So
10:53, what would AI really change?
10:57Well, perhaps the essential, because
10:58there is a fundamental difference
11:00between being able to bend a government
11:02and becoming indispensable to the very
11:03exercise of its power.
11:06Take Google or Facebook. Their power to
11:09pressure states is a secret to no one.
11:12In 2021, Australia wanted to force
11:14platforms to pay the press.
11:16Google threatened to withdraw its search
11:17engine from the country.
11:19Facebook, for its part, took action.
11:21Overnight, there were no more news
11:22articles on the platform for Australians
11:24.
11:25Even public health agency pages were
11:27blocked. All this just days before the
11:29launch of the Covid vaccination campaign
11:31.
11:32And guess what?
11:33Without hesitation, the government
11:35agreed to amend its law and Facebook
11:36restored the news. It's a real power
11:39struggle, yes, but the pressure comes
11:41from the outside. The company threatens
11:43to cut a service to citizens and the
11:45state negotiates with its own means.
11:48It can fold like Australia, but it can
11:50also hold firm.
11:51In Canada, Meta has been blocking news
11:53since 2023 and Ottawa hasn't changed its
11:55law.
11:56With AI, the leverage would no longer be
11:59outside the state, in the services
12:01citizens use, but inside, in the tools
12:03with which the state itself thinks and
12:06acts.
12:07The minister still chooses, certainly,
12:09but if their teams can no longer produce
12:11independent analysis or if it takes them
12:13a week to accomplish what a model does
12:15in 5 minutes, what alternative do they
12:17really have left?
12:20You can have the final say while losing
12:21control over the options you are
12:23choosing between.
12:25AI also introduces a risk of an almost
12:28historic reversal.
12:29In 1945, de Gaulle created the Atomic
12:32Energy Commission.
12:34A year later, the United States went so
12:36far as to write into law that
12:37information regarding nuclear weapons is
12:39classified from its creation, even when
12:41produced outside the state.
12:44The atomic bomb, the first networks that
12:46would lead to the internet, the GPS—so
12:48many strategic technologies developed
12:50within major public programs with the
12:52participation of private companies,
12:54certainly, but always driven by the
12:56state.
12:57The state didn't just regulate these
12:59technologies.
13:00It directed their development and
13:02largely possessed knowledge of the
13:04danger.
13:06But with AI, it's the opposite. In 2025
13:08, more than 9 out of 10 reference AI
13:11models came from the private industry.
13:14This time, the state must regulate a
13:16technology whose most advanced
13:17capabilities are developed primarily
13:19outside of it in the private sector.
13:22And even those who build them admit it.
13:24They do not fully understand what they
13:26are doing. They are not capable of
13:28precisely explaining all the internal
13:30mechanisms of their own models. And I am
13:32quoting Dario Amodei, the head of
13:34Anthropic. This ignorance is virtually
13:36unprecedented in the history of
13:37technology.
13:39With each passing quarter, models become
13:42more powerful, more useful, and
13:43integrate a bit more deeply into our
13:45societies.
13:47And if no one understands them perfectly
13:49, some still have a considerable
13:51advantage over others: those who build,
13:54train, and test them.
13:56So, I ask you, a government that uses
13:59these tools for essential functions,
14:01while fearing it no longer knows how to
14:04measure the risks, who will it turn to?
14:07We can always tell ourselves stories by
14:08saying that states are creating
14:10institutions to evaluate these systems.
14:12Okay, but look at the numbers.
14:14The British AI Security Institute, one
14:17of the best funded in the world, has a
14:19few tens of millions of pounds per year
14:21.
14:22Its American equivalent, about fifteen
14:25million dollars. Last year, private
14:27investment in AI in the United States
14:30exceeded 285 billion dollars. A ratio of
14:331 to nearly 20,000.
14:36So obviously, a regulator does not need
14:38to spend as much as the companies it
14:40oversees, but when those companies
14:41concentrate the models, the
14:43infrastructure, the researchers, and a
14:45large part of the means to understand
14:46them, how can you hope to compete in
14:48terms of expertise and evaluation
14:50capacity?
14:52The state would depend on the same
14:54companies to govern with AI and to know
14:56how to govern AI.
14:58This is the most subtle trap in this
14:59story. All of this would be perfectly
15:01rational.
15:02The best reason to entrust them with
15:04this power is also the best reason to be
15:06worried about it.
15:07And this power could grow even further
15:10if, at the same time, automation
15:12weakened the power that work gives us.
15:18Because we do not have influence in
15:20society solely because we vote. We also
15:22have influence because we are needed.
15:24Companies need employees,
15:25administrations need skills, and the
15:27state needs an economy that produces
15:29income it can tax.
15:31It is this need that gives weight to our
15:34actions and our demands. A strike,
15:36fundamentally, is nothing more than that
15:38. Reminding someone that they need you.
15:40You will tell me that protesting or
15:42striking rarely changes much. That is
15:44not entirely false. But this power is
15:46not only played out in the streets.
15:48Typically, after Covid, American
15:50employers were short on labor.
15:53Well, low wages grew faster than all
15:55others, to the point of erasing about a
15:57third of the rise in wage inequality
15:59accumulated since 1980.
16:01And that was made possible thanks to
16:03employees who switched employers to get
16:06better pay, simply because they were
16:08needed. Of course, that doesn't explain
16:10everything, obviously, but it does
16:12remind us of a simple rule. We get more
16:14from those who need us. Now, look at
16:17countries where the power does not
16:19depend on its population for funding. In
16:21an oil state, for example, the
16:23government doesn't live off taxes from
16:25its citizens; it lives off rent. And so
16:27, very often, it has less need for their
16:29contribution and holds itself less
16:31accountable to them.
16:32We call this the rentier effect, one of
16:34the manifestations of the resource curse
16:36.
16:37Luke Drago and Rodolph Lane, two
16:39researchers, asked themselves the same
16:41question regarding AI.
16:42What if it produced the same effect, but
16:45everywhere?
16:47Wealth produced with less and less human
16:49labor, held by a few players, would risk
16:51weakening the incentives to invest in
16:53the rest of the population.
16:55They call this the intelligence curse.
16:58In other words, as soon as an increasing
17:00share of wealth is produced by machines
17:02concentrated in a few hands, the
17:04influence of those who own them grows to
17:06the detriment of the influence of
17:07workers, which diminishes. Add to that
17:10everything we have just seen.
17:12The same players would provide the state
17:14with its tools to govern and an
17:15increasing share of its revenue to
17:17function. And this is where you're going
17:19to tell me, "Anyway, Richard, there will
17:21be a guaranteed universal income." So we
17:23don't care. Very well. But guaranteed by
17:25whom?
17:26By the state. So, it would be funded by
17:28the profits of these few companies. Your
17:31income would depend on their results and
17:33the state's ability to tax them. We fall
17:35back into exactly the same dependency,
17:37even if it were guaranteed by the
17:39companies themselves. For one, I am
17:41curious to see how they will justify
17:43that in the name of independence and
17:44democracy—even superficial democracy—
17:46because, regardless, they are inventive
17:47. But in exchange for what? A salary is
17:50an exchange. You give your labor, and
17:51you can take it back. That is what gives
17:54you leverage. A handout is not in
17:55exchange for anything, or else it is for
17:58conditions you did not set.
18:00We can go on strike, we can play off the
18:01competition between employers. Generally
18:03, one does not go on strike against an
18:05allowance.
18:06We could therefore find ourselves living
18:08better materially speaking, while having
18:10less influence on the organization of
18:11society.
18:12It is, in a sense, the "labradorization
18:14" that Musk talks about. My dog lives
18:17better materially with me than he would
18:19if he were all alone in the wild. On the
18:21other hand, he has far less control over
18:23the decisions that shape his existence.
18:26Obviously, we are still far from all of
18:27that, and it is not about being
18:29deterministic. What interests me is the
18:31trajectory. Yet, some people are not
18:33just considering this trajectory. They
18:35are announcing it, and sometimes they
18:37are even calling for it.
18:39In November 2025 in Washington, at the
18:41American Investment Forum, Elon Musk
18:43predicted that within 20 years, work
18:45will be optional.
18:47Less than a year later, in Paris, he was
18:49already mentioning a high universal
18:51income to allow everyone to live without
18:53working.
18:54Well, on the question of who will fund
18:56it and under what conditions, Musk
18:58remains much less precise.
19:00But while one person announces the end
19:01of work, another worries about the power
19:03of those who would want to regulate the
19:05technologies that could replace it.
19:07Since September 2025, Peter Thiel,
19:09founder of Palantir, has been appearing
19:11at conferences all over the map to
19:13defend his thesis.
19:15The fear of existential risks,
19:17especially those linked to AI, could
19:18become the pretext for establishing a
19:20global government with totalitarian
19:22tendencies.
19:23And in his discourse, some of those
19:25warning about technological risks become
19:27, and I quote, "legionnaires of the
19:29antichrist." Yes, you heard that right,
19:31and it is not a recent whim.
19:33As early as 2009, he wrote, "I no longer
19:35believe that freedom and democracy are
19:38compatible."
19:39The guy is basically bringing back
19:41Plato's critique of democracy 2,400
19:43years later.
19:45Notice the paradox: we are here
19:47discussing a future where machines could
19:49replace a large part of human labor
19:51while some of their most powerful
19:52promoters warn us against the political
19:55power that might seek to control them.
19:58It is fascinating. Everyone accuses
19:59everyone else of using fear to
20:01accumulate power.
20:03And I could have continued with Sam
20:05Altman, who wrote back in 2021 that more
20:07power will shift from labor to capital.
20:10In short, this is precisely where the
20:12problem lies. A society where we no
20:15longer need to work could be a society
20:17of abundance.
20:18Yes, but it could above all become a
20:20society where those who produce wealth
20:23no longer need to listen to others
20:25either.
20:26So no, all of this does not constitute
20:28an absolute certainty of the direction
20:30we are taking, obviously. But this
20:32future is no longer just a scenario
20:34imagined by a few crazy minds.
20:37It is being announced, debated, and
20:39sometimes desired by those who have the
20:41very means to make it possible and who
20:43could achieve it through the meeting of
20:46companies seeking to become
20:47indispensable and governments seeking
20:49immediate results.
20:55At the start of this video, I told you
20:57that all that was missing to make a
20:59state was the logo on the passport. In
21:01reality, these companies will never need
21:03one. They don't have to overthrow the
21:06state, or even replace it. They just
21:08need to become the ones without whom it
21:10can no longer understand, decide, or act
21:13.
21:13States would keep everything you can see
21:15: the flag, the elections, the law, the
21:17taxes. They would lose everything you
21:19can't see: the capacity to do otherwise
21:21. AI could therefore make states more
21:23efficient while making them less
21:25autonomous, until the moment where,
21:27lacking alternatives, a government must
21:29adapt the program chosen by its voters
21:32to the conditions of its suppliers. And
21:34what if none of this is set in stone? Of
21:36course, the only thing that is certain
21:38is that we won't stop AI.
21:40If you want my opinion, the real answer
21:42is competition, distribution,
21:44decentralization, and in the best of
21:46worlds, open technologies.
21:48Ensuring that every player can create
21:50value, but that no one becomes essential
21:52and that we can always choose to leave.
21:55Because a state that can no longer
21:56change suppliers doesn't really have a
21:58supplier anymore. It has a guardian.
22:01And if you want to learn more about the
22:03convergence of Big Tech and the deep
22:05state, go watch this video on Alexander
22:06Karp from Palantir. Yeah.

About this transcript

This page contains the full transcript of La Maison-Blanche cache quelque chose dont personne ne parle by Grand Angle, generated from the public captions YouTube serves with the video. The transcript has 3,791 words across 651 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.