Trump, the AI chiefs, and the strange promise of corporate self-control. The Machiavellian Playbook of AI.
On September 29, 2026, Donald Trump sat down at the White House with some of the most powerful people in artificial intelligence.
Sundar Pichai of Google was there. So were Dario Amodei of Anthropic, Mark Zuckerberg of Meta, Greg Brockman of OpenAI, Elon Musk of xAI, and Jensen Huang of Nvidia. Together they signed the White House Accord on Super Intelligence.
The scene perplexed me.
For years, people inside the AI industry have warned us about systems that could manipulate human beings, threaten cybersecurity, assist with biological weapons, or become too difficult to control. The language sometimes approaches the apocalyptic.
Then many of the same companies stood beside a president known for opposing strong regulation and agreed that they should largely police themselves. The message seems to be:
“Artificial intelligence may become a danger to humanity.
But please leave the supervision to us.”
That contradiction deserves a closer look.
What the accord actually says
Let us begin with the facts.
The accord asks companies developing frontier models to establish four layers of control:
- Internal systems to monitor dangerous capabilities.
- An internal team to check whether those systems work.
- An independent external auditor or evaluator.
- A committee of the company’s own board to receive reports and oversee repairs.
The participating companies also promise to meet regularly and develop common safety standards. These measures are not meaningless. Cybersecurity testing, outside evaluation, and board responsibility can all reduce risk. A company building powerful systems should have them.
But we must also notice what the document does not contain. It creates no statutory penalties. It requires no public disclosure of audit findings. It names no government regulator with inspection powers. It gives citizens no legal remedy if the controls fail. Even the accord leaves the selection of the outside evaluator to each company.
The accord ends by saying that “over time, it may make sense” to put these measures into law. Until that unspecified time, compliance remains voluntary.
Those are facts visible in the text of the accord itself.
On the same day, Trump signed a separate executive order directing the federal government to replace the term “Artificial Intelligence” with “Super Intelligence.” The order describes the technology mainly through the language of promise, national leadership, and economic opportunity. It also states that it creates no legally enforceable right or benefit. That too is part of the public record. (White House executive order)
Now the interpretation begins.
Safety without surrendering control
The accord may be read as a safety document. I’d rather read it as a document about power.
The companies accept the language of responsibility, but they do not surrender much authority. They monitor their own models, appoint their teams, select outside evaluators, and report to their boards.
The builders remain in the driver’s seat.
This is the attraction of voluntary self-regulation. It allows a company to appear responsible without accepting the harder parts of public law: compulsory disclosure, government inspection, civil liability, fines, or a public authority able to order a stop.
The government can say it acted. The companies can say they are safe. Development continues.
We should be careful here. The text does not prove a secret conspiracy. We do not need one to understand what is happening. Institutions usually protect their freedom of action. Companies protect their markets. Political leaders prefer agreements that produce a strong image without creating a difficult enforcement system.
The result is Machiavellian, even when every participant believes he is acting reasonably.
Machiavelli understood that political power depends partly on appearance. A ruler must be considered careful, moral, and necessary. Modern corporate power works in a similar way. The company must appear innovative but responsible, powerful but trustworthy, fast but under control.
The accord provides that appearance.
Fear can become a business advantage
There is another uncomfortable question.
When an AI company warns that its own technology could become extraordinarily dangerous, the warning may be sincere. Researchers can genuinely fear loss of control, cyberattacks, biological misuse, or systems acting beyond their intended purpose.
But the same warning can also strengthen the company.
If advanced AI is described as almost too dangerous for ordinary people to handle, then only the largest laboratories appear qualified to build it. They have the computing power, the safety departments, the lawyers, and the money to meet complex standards. Smaller companies and open-source developers do not.
Existential-risk language can therefore serve two purposes at once. It can express a real concern and create a competitive wall around the biggest firms.
That does not make every warning dishonest. It means we should follow the incentives as well as the words.
Who gains authority when the technology is described as nearly godlike? Who gets invited to the White House? Who helps write the safety rules? And who is told to trust the people already building the machine?
The steering gap
In my earlier essay The Steering Gap, I argued that the deepest problem of AI is not raw intelligence alone. It is the distance between machine capability and human control.
A tool waits for instruction, but an agent can act!
It can read files, send messages, spend money, recommend medical choices, evaluate workers, negotiate with insurers, or influence a child. Once systems act in our name, governance cannot stop at monitoring their output.
We must ask who gave the system authority, what limits were set, who can withdraw that authority, and which human institution is liable when something goes wrong.
The White House accord does not answer these questions. It treats governance mainly as a chain of corporate controls.
That is precisely where caution is needed. Procedures are useful, but procedures can also become a ritual. A safety policy, an audit report, and a board committee may create the appearance of control while decision-making remains hidden from the public.
The steering wheel is still inside the company. Society is asked to trust that someone responsible is holding it.
Apocalypse and rescue
The White House scene also reminded me of my earlier essay on Peter Thiel, Armageddon, or Antichrist?
Thiel has warned that fear of catastrophe can prepare people to accept new forms of control. The promised rescuer may become the ruler. Emergency language makes ordinary limits look irresponsible.
His warning should also be applied to the technological elite itself: what happens when private companies describe an approaching technological catastrophe and then present themselves as the only people capable of preventing it?
They become both the makers of the danger and the providers of rescue.
This does not mean the danger is imaginary. Nuclear dangers are real, even though governments also use them politically. AI risks can likewise be real while the language surrounding them serves power.
The important question is who receives authority from the fear. In the White House accord, the answer is clear enough. The companies receive public recognition as responsible guardians, while binding public oversight remains weak and future regulation is postponed.
The apocalypse is announced. The rescue is voluntary. Business continues.
The technological system protects its movement
Theodore Kaczynski made a severe argument about industrial society: once a technological system becomes large enough, society begins adapting human life to the needs of the system. Technology stops being a collection of optional tools. It becomes an environment people cannot easily leave.
Kaczynski was a terrorist and murderer. Nothing in his violence should be excused. His demand for the collapse of industrial society is not my conclusion.
But an argument does not become false only because its author became monstrous.
His critique of voluntary ethical codes remains disturbing. A system driven by competition, investment, national rivalry, and technical possibility will find it very difficult to limit itself. Each company may say that slowing down would hand the advantage to another company. Each country may say that restraint would hand leadership to another country.
Everyone can understand the danger. No one believes he can afford to stop. The result is acceleration presented as necessity.
A voluntary accord fits this system perfectly. It absorbs criticism without changing the direction of travel. Safety becomes another department inside the machine.
What real responsibility would require
I do not believe the answer is to stop all AI development. Nor do I believe that every internal safety effort is a fraud. The question is simpler: should the institutions creating a potentially society-changing technology also have the final authority to decide whether their precautions are enough?
I do not think so.
Real responsibility would require clear legal duties for systems with serious effects on health, work, credit, education, security, and political life. It would require audit results that public authorities can inspect. It would preserve a meaningful right to refuse automated decisions. It would keep liability attached to companies and people rather than allowing it to disappear inside technical systems.
Most of all, it would give someone outside the company the legal power to say no. Without that power, “oversight” remains advice.
The White House accord tells us that its signatories want the public to trust them. Perhaps some of them deserve a measure of trust. But democratic government was not invented because every powerful person is wicked. It was invented because goodwill is not a sufficient control on power.
A company may sincerely want to protect humanity and still protect its market position first. A president may speak about safety and still prefer spectacle to enforcement. An auditor may be called independent while depending financially on the industry being audited.
These are not personal accusations. They are ordinary problems of institutional design.
The document signed in Washington offers four layers of control. Yet all four remain close to the companies themselves.
So I return to the question that has followed me through my writing on artificial intelligence:
Who remains able to steer?
Not who attends the meeting. Not who signs the pledge, and not who promises to behave responsibly. Who can stop the system when stopping becomes costly?
Until we have a convincing answer, the public is not holding the steering wheel. We are sitting in the back seat, listening to the builders assure us that the brakes have been checked.
Good luck, Humanity!
‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑
© Robert F. Tjón, October 2026
Creative Commons CC BY-NC-ND 4.0 International
Video: “The Machiavellian Playbook of AI” on
This piece first appeared on Substack. I republish it here voluntarily—not as repetition, but as a trace; a place where words can rest after their first flight.
Legend
- AGI: Artificial General Intelligence.
- AI: Artificial Intelligence.
- AI Act: The European Union Artificial Intelligence Act (Regulation EU 2024/1689).
- FC: Freedom Club (Theodore Kaczynski’s pseudonym).
- LGPD: Lei Geral de Proteção de Dados (Brazilian General Data Protection Law).
- OpEx / CapEx: Operating Expenses / Capital Expenditures.
Sources & References
https://www.presidency.ucsb.edu/documents/white-house-accord-super-intelligence
https://www.youtube-nocookie.com/embed/k1sAOxPXt70?rel=0&autoplay=0&showinfo=0&enablejsapi=0
The Steering Gap: What the new AI Laws are really trying to protect
Tjón, R. F. (2026). Armageddon or Antichrist? Why one Silicon Valley billionaire now frames the future in openly apocalyptic terms. https://rftjon.substack.com/p/armageddon-or-antichrist?r=35vtu2
Tjón, R. F. (2026). The Architecture of Humanity: What AI Reveals About the Systems We Built. https://rftjon.substack.com/p/the-architecture-of-humanity?r=35vtu2
Kaczynski, T. J. (1995). Industrial Society and Its Future. The Washington Post. https://www.washingtonpost.com/wp-srv/national/longterm/unabomber/manifesto.text.htm
AI and the modern Frankenstein — and who the Real Monster is
There are stories that stretch like shadow lines across centuries. Mary Shelley’s Frankenstein is one of them.
AI disclosure: I use artificial intelligence to create visuals and for data synthesis and structural mapping. AI-generated insights may contain inaccuracies; readers should verify critical information independently. All conclusions and final editorial decisions remain the original work of Robert F. Tjón.
This content is licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International.
