What the Financial Times actually revealed
According to the Financial Times, Sam Altman has discussed a plan giving the US government a 5% stake in OpenAI, with no financial consideration in return. On the basis of OpenAI’s valuation in March 2026, estimated at 852 billion dollars, that share would represent around 42.6 billion dollars — so the figure of 42 billion cited in the video is near enough exact.
The plan was discussed with Donald Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent. Altman also raised it separately with Senator Bernie Sanders. And it does not stop at OpenAI: Altman reportedly wants Anthropic, Google and Meta each to hand over an equivalent stake, which corresponds exactly to the video’s wording — “5% of its big American labs”, in the plural.
An important point the video does not mention: setting up the fund would require a vote by the US Congress. For now, nothing is settled. TechRadar published a corrective piece on precisely this point, headlined in substance “a stake for the state does not mean a stake for you”: the distribution mechanics — who gets what, when, in what form — remain entirely to be defined.
The Alaska model, a fund dating from 1976
The comparison with Alaska, on the other hand, holds up perfectly.
Alaska’s Permanent Fund was created in 1976 to capture a share of the oil revenues extracted on the state’s soil. Since 1982, every resident has received an annual cheque — the first payments were 1,000 dollars. The principle: a locally extracted resource funds a dividend paid exclusively to local residents.
Transposed to AI, Altman’s reasoning amounts to saying: the large models are a strategic resource comparable to oil, the American state captures a share of it, and American citizens receive the dividend.
This is not a new idea, either. As early as March 2021, in an essay titled “Moore’s Law for Everything”, Altman was already proposing an “American Equity Fund”, financed by a tax of 2.5% a year on the market value of large companies and 2.5% a year on the value of privately held land, with an annual distribution to every citizen over 18 — projecting around 13,500 dollars per adult per year within a decade or so. His formula at the time: “The new social contract will be a floor for everyone in exchange for a ceiling for no one.”
Five years later, that same man runs the company that weighs most heavily in that equation. What was a theoretical essay becomes a concrete proposal, negotiated with the White House.
Why this cheque does not cross the Atlantic
Here is the point the video brushes past without developing, and which should nonetheless be the heart of the matter.
This fund, if it comes into being, will be financed by American companies, approved by an American Congress, and — if the Alaska model is followed through — paid to American residents. Full stop.
A user in Dakar paying 20 dollars a month for ChatGPT Plus is also financing OpenAI’s growth. A freelancer in Casablanca using Claude for their work is also generating data and value. A civil servant in Brussels who depends on a tool built on these models is also subject to decisions taken in Californian boardrooms. None of the three will ever receive a cent from this fund. Not through oversight. By design.
Senator Sanders is pushing a far more ambitious alternative version — a one-off tax of 50% on the shares of large AI companies, to build an “AI sovereign wealth fund” estimated at 7,000 billion dollars, with an annual payment of more than 1,000 dollars per American. Whether the Altman version or the Sanders version wins out in Congress, the result is the same on one point: both plans are designed for American citizens.
This is not a drafting accident. It is the Alaska model applied to the letter: one resource, one territory, one eligible population. And that population does not cover the 7.8 billion people who, elsewhere in the world, already use those same tools.
Konwinski: “feudalism, with better branding”
The video then cites “two figures who study the concentration of power in AI”. The first is Andy Konwinski, co-founder of Databricks and co-founder of Perplexity.
The exact quote, reported by journalist Alex Heath in his newsletter Sources on 19 June 2026, is not quite the one in the video. Konwinski does not talk about “marketing” but about “branding” — a machine-translation detail, but one worth correcting: “Feudalism, with better branding. Two or three companies end up owning the fundamental resource, while everyone else rents access on their terms.”
The 100 million dollars mentioned do not date from this week. Konwinski committed them back in June 2025 to found the Laude Institute, dedicated to open AI research — the video presents that figure as news, when the institute was celebrating its first anniversary at the very moment of these statements.
What is genuinely new, however, is the escalation of the last three weeks. On 9 June 2026, Anthropic launched Claude Fable 5 and its more powerful model, Mythos 5 — and the model card revealed that it was silently degrading its answers to users suspected of training a competing AI. That episode becomes Konwinski’s central example. On 30 June, he gathered around a hundred researchers at San Francisco’s Exploratorium, at a meeting called “Open Frontier”. On 2 July, he published an essay titled “The concentration of power in AI is a risk, not a solution”, with this line: “The problem is not that Anthropic made a bad decision. The problem is that it assumed the decision was theirs to make.”
LeCun: an old departure, a real warning
The second name cited by the video is Yann LeCun, presented as one of the three “fathers” of modern AI, alongside Geoffrey Hinton and Yoshua Bengio.
Here, the video misleads on one specific point: it presents LeCun’s departure as news from this week. In fact that departure from Meta was confirmed on 19 November 2025 — eight months earlier. Since then, LeCun has co-founded AMI Labs in Paris, where he holds the post of executive chairman (the chief executive role goes to Alexandre LeBrun). The company develops “world models” based on the V-JEPA architecture, an alternative to classic large language models, and raised 1.03 billion dollars in March 2026, at a pre-money valuation of 3.5 billion dollars.
What is genuinely new is his public reaction to Konwinski’s essay. On 3 July 2026, LeCun replied on X: “The concentration of power in AI and the desire for control are, by far, the greatest danger of AI.” He reuses a comparison he has in fact been using since 2023 — the Ottoman Empire, which banned the printing press for two centuries to protect the power of the clergy and the calligraphers’ guild. The video presents that analogy as a find of the moment; it is in reality an argument LeCun has been recycling for several years. That does not make it false. It only makes it less spontaneous than the video suggests.
Konwinski and LeCun did not write a joint document, contrary to what the video implies. There is an essay published by one, and a public response from the other — a public convergence, not a formal coalition.
Mythos, not Meta: the real name behind the export control
The most muddled passage in the video concerns an export block negotiated “after 1 July”. Once the badly transcribed names are corrected, the story is clear — and we had already documented it.
On 12 June 2026 at 23:21 Paris time, the US government ordered Anthropic to cut off access to Fable 5 and to its most powerful model, Mythos 5, for every foreign national in the world, including Anthropic’s own foreign employees. We told that episode in our article on the suspension of Fable 5.
On 26 June, a first relaxation arrived: Commerce Secretary Howard Lutnick authorised access to Mythos 5 for around a hundred American organisations “that operate and defend critical infrastructure”. Fable 5, the consumer model, stayed blocked. Then, between 30 June and 1 July, a second relaxation lifted export control on both models entirely, in exchange for a commitment from Anthropic to cooperate more closely with Washington: a shared framework with Amazon, Microsoft and Google, early government access to future models judged sensitive for national security, and joint testing.
No source mentions Meta in this episode. The exact name, in every newsroom checked, is “Mythos” — the confusion with “Meta” is almost certainly an artefact of automatic transcription. But the substance of the affair holds up perfectly: for two weeks, access to the world’s most advanced model in cybersecurity terms was decided entirely by a handful of American officials, with no foreign government having any say.
What the video is right to point at
Once the transcription errors are corrected, the underlying message survives verification: a handful of American actors — a president, two cabinet secretaries, a Commerce regulator, three or four AI labs — is deciding, right now, who has the right to use which artificial intelligence, and who will receive a share of its value. The rest of the world is associated with none of those decisions. It inherits them.
What the video simplifies or distorts
Five points deserve correction: the 42 billion dividend has not been voted, only proposed; LeCun’s departure is not July news, but November 2025; Konwinski’s 100 million dates from June 2025, not from this week; “Meta” is very probably a mistranscription of “Mythos”; and the Ottoman Empire analogy is not new, LeCun has been repeating it since 2023. None of these corrections invalidates the overall message. They only avoid presenting as sudden news what is, for a good part, a file that has been building for months.
What it changes, concretely
For an American resident, the sequence of these past weeks sketches a coherent scenario: a potential dividend indexed to the value of the AI labs, priority access to the most advanced models as soon as they touch national security, and real decision-making power over what will or will not be exported. Nothing is guaranteed — Congress still has to vote — but the balance of power is written: the resource, the control and the potential benefit all stay on the same territory.
For a user located elsewhere, the same sequence reads in reverse. Same subscription paid, same dependence on American tools, same exposure to the risks documented by Konwinski and LeCun — concentration of knowledge, compute and electricity in a few hands — but no access to the resource, no voice in the regulation, and access to the most powerful models that can be cut off overnight by a simple directive, as happened on 12 June. The European AI Act, supposed to offer a regulatory counterweight, has itself just been weakened and postponed under the joint pressure of the American giants and the Trump administration — which further reduces the rest of the world’s capacity to weigh on these decisions.
One way out remains, which the video is right to mention in conclusion: open models, downloaded and run locally. No passport to present, no vote of Congress to wait for, no conditional dividend to hope for — and no government directive can switch them off remotely. That is precisely the resource Konwinski is seeking to fund with his institute, and one of the rare levers that escapes entirely from the logic described above.
Le Recul’s reading
None of these four episodes is isolated. They happened within the same four-week window, between 9 June and 3 July 2026: the launch and then the blocking of Mythos 5, the lifting of that block in exchange for closer cooperation with Washington, the public warning from two respected AI figures about the concentration of power, and a national dividend proposal that resurrects an essay five years old. This is not a coincidence of the calendar. It is the same dynamic expressing itself in four different forms: the resource, the control and the benefit stay on the same side of the Atlantic, while the use, the dependence and the risk are spread everywhere else.
So the question to ask with every new announcement about “sharing AI” is not whether the gesture is generous. It is where the cheque actually goes — and who, in the room where the decision was taken, had the right to be there.
What to take away
OpenAI has proposed handing 5% of the company to the US government, a stake estimated at 42.6 billion dollars, revealed on 2 July 2026. The plan takes its cue explicitly from Alaska’s Permanent Fund and requires a vote by Congress.
Andy Konwinski, co-founder of Databricks and Perplexity, has been warning since June 2026 about “feudalism with better branding” in AI — his 100 million dollar funding for open research dates from June 2025.
Yann LeCun, who left Meta in November 2025 and now heads AMI Labs, publicly described the concentration of power in AI as the sector’s “greatest danger”, on 3 July 2026.
Between 12 June and 1 July 2026, access to Anthropic’s Fable 5 and Mythos 5 models was blocked, then partly and then fully restored, purely by decision of the US government.
The figure to remember
0.
That is, to date, the guaranteed share of this AI dividend for anyone living outside the United States — whatever their usage, their subscription, or their real contribution to the value of these models. Not through a calculation error. By design.