Skip to content

Before the crisis breaks

Who controls the people building AI? Sam Altman and a crisis that communication cannot solve

Sam Altman says that no one person, company or country should control a technology this powerful. That leads to an uncomfortable question: who controls the people who build it?

Sam Altman warns of the power of AI.

He talks about risk.

About safety.

About the need for control.

And at the same time he heads one of the organisations developing this technology the fastest.

This is not an accusation.

This is the paradox of power.

Because the more powerful artificial intelligence becomes, the less adequate the answer:

“trust the people who build it”.

We need an answer to a much harder question:

who controls them?

That question is particularly topical today.

An OpenAI safety employee has just left the company and publicly called its safety culture into question.

The largest AI companies have been summoned to testify in public, under oath, before the New York City Council.

Sam Altman’s conflict with Elon Musk has long gone beyond a business disagreement.

And Luca Guadagnino’s “Artificial” is opening in cinemas — a film that revisits a few extraordinary days in 2023, when the OpenAI board fired Altman, employees threatened to leave en masse, and within days the CEO was back, stronger than before.

A governance crisis has become pop culture.

But the problem it exposed has not gone away.

When the leader becomes the brand

For years, Sam Altman’s personal brand was an important asset to OpenAI.

Calm. Rational. Speaking in the same breath about AI’s enormous potential and the need to control its risks.

It is a remarkably effective reputational construct.

Such a leader is selling more than a product.

He is also selling a belief:

“we know the power of what we are building — and that is why you can trust us to be careful”.

The problem begins when the leader becomes almost synonymous with the organisation.

Because then his reputation no longer belongs to him alone.

His conflict becomes the company’s conflict.

His credibility affects the credibility of its commitments.

His opponents become the brand’s opponents.

And questions about his motives begin to turn into questions about the motives of the entire organisation.

This is classic key-person reputation risk.

Only on an extraordinary scale.

One sentence can change the whole narrative

Altman’s latest remarks show how difficult this problem has become to manage in communication terms.

On the one hand, he speaks of the need to take the risk posed by superintelligence very seriously and admits that the alignment problem — ensuring that ever more powerful systems behave in line with human intentions — has not been solved.

On the other, he argues that society will probably have to accept a certain level of negative consequences from AI, because the benefits of broad access to the technology may be far greater.

These positions need not be logically contradictory.

Altman distinguishes between limited harms and catastrophic risk.

But communication does not end with the sender’s intention.

The audience may hear:

“we develop the technology, and society will bear part of the cost of developing it”.

And then an even harder question arises:

who, exactly, was given a mandate to decide what level of risk society should accept?

This is where a technological problem becomes a problem of reputation and governance.

The toughest critic comes from within

On 3 October, Reuters reported the departure of David Robinson, an OpenAI employee who worked on safety.

Robinson said publicly that, in his view, the company’s culture is not cautious enough given the pace at which ever more powerful AI systems are being developed, and that the time for trial and error is over.

That is his assessment.

Not proof that OpenAI ignores safety.

But in reputational terms, the source of an allegation matters enormously.

Criticism from a competitor can be put down to market rivalry.

Criticism from a journalist — to interpretation.

Criticism from a regulator — to institutional pressure.

But when the warning comes from someone working inside the safety system, it strikes at the very heart of the organisation’s credibility.

And that is why high-risk companies need more than a whistleblowing policy on paper.

They must be able to show that inconvenient information can reach a place where it will be assessed independently of hierarchy and business interests.

Musk versus Altman. When governance becomes a personal conflict

Elon Musk’s conflict with Sam Altman has been going on for years.

It concerns OpenAI’s history, its mission, structure, commercialisation, the direction of its development and the vision of AI’s future.

Over time, it has moved into the courts, interviews and social media.

For the public, it is a spectacle.

For corporate affairs, it shows something more important:

what happens when a governance dispute becomes personal.

Two extremely high-profile technology leaders are publicly questioning each other’s decisions and intentions.

Yet AI safety may require exactly the opposite: cooperation between competitors, shared standards, the exchange of information about serious incidents, and agreement on the lines that should not be crossed even in the midst of a technology race.

At that point, the personalisation of the conflict stops being merely an image problem.

It can become a systemic problem.

Regulators are starting to ask a different question

On 5 October, representatives of the largest AI developers appeared before the New York City Council at a public hearing on artificial intelligence safety.

The issues on the table include independent validation of systems, protection for those who report risks, incident-related obligations and liability for foreseeable harm.

That is an important shift in language.

For several years, the question was:

“can we trust the companies developing AI?”

Increasingly, it is:

“what evidence of safety should we demand of them?”

That is a fundamental difference.

Because the reputational formula:

“trust us”

is starting to give way to the institutional one:

“show us how you control it”.

OpenAI has an answer. The question is whether it is enough

It would be unfair to tell this story as if OpenAI were not building governance mechanisms.

It is.

Following the restructuring, the OpenAI Foundation controls OpenAI Group PBC and holds powers over appointments to its board.

The Foundation Board has a Safety and Security Committee overseeing the organisation’s safety and security practices.

In September, Paul Christiano, who has worked on the safety of advanced AI systems for years, joined the Foundation Board. He also took a seat on the Safety and Security Committee.

OpenAI is also developing its Frontier Governance Framework, which covers, among other things, cyber risks, CBRN, manipulation, loss of control, incident response and the involvement of external experts.

These are real mechanisms.

So the right question is not:

“is OpenAI doing anything about safety?”

It is.

The question is:

is governance keeping pace with the technology, the business and the potential consequences of a mistake?

PR will not fix this crisis

In a classic crisis, a company can improve its communication.

Explain.

Apologise.

Change its spokesperson.

Publish a report.

Here, that will not be enough.

If the problem is:

“who controls the technology, and who controls the people who build it?”

the answer cannot be:

“you can trust us”.

A far stronger answer is:

“you do not have to rest safety on trust in one person. Here is a system that controls that person too”.

And this is precisely where reputation meets governance.

How can this be prevented?

I see five safeguards.

1. Genuinely independent oversight

Safety cannot be subject solely to the same incentives that reward speed of development, competitive advantage and business performance.

2. A safe escalation path

An employee who identifies a serious risk must be able to go outside their own reporting line without fear that challenging a business decision will end their career.

3. External validation

A company should not be the only party assessing the safety of a system it builds itself.

But that raises a further question:

who controls the controllers?

The auditor, too, needs independence, competence and safeguards against conflicts of interest.

4. Transparency about incidents and near misses

Trust does not come from an organisation insisting that it makes no mistakes.

It comes when stakeholders see how mistakes are detected, escalated and corrected.

5. Separating the CEO’s personal brand from the system of trust

A charismatic leader can be an enormous asset to a company.

They cannot be its only mechanism of credibility.

Because reputation shows up on the balance sheet very quickly

First comes the question about the leader.

Then about the organisation’s culture.

Next, about product safety.

After that, about the effectiveness of oversight.

If the answers fail to convince, regulatory pressure follows.

In its wake may come new obligations, higher compliance costs, legal risk, caution among partners and investors, and an ever harder fight for the social licence to keep developing the technology.

And then something happens that companies sometimes forget:

reputation stops being the communications department’s problem.

It becomes a cost of doing business.

The most important lesson is not about Sam Altman

“Artificial” may mean that, over the coming weeks, he is the one we talk about.

Whether he is a visionary.

Whether he is a ruthless operator.

Whether the OpenAI board was right in 2023.

Whether Musk is right.

The film may offer its own answer.

The media — their own.

Altman’s opponents — yet another.

But from the standpoint of reputation management, these questions are secondary.

A safe system should not require society to take it on faith that the person at the top is a good person.

It should work regardless.

That is why the most important question is not:

“can we trust Sam Altman?”

Rather:

why should the safety of a technology that could change the world depend at all on whether we trust Sam Altman?

And that is a lesson far broader than AI.

The best way to prevent a reputational crisis is not always better communication.

Sometimes it is a better-designed organisation.

Sources and further reading

First published in Polish on LinkedIn, 7 October 2026.

More analyses

Before the crisis breaks

“It was not mine.” Ed Sheeran is left on his own — but can reputation be handed to a promoter?

Macklemore was dropped from the tour. Then every announced support act withdrew, and Beoga, who played during part of Sheeran’s own set, also pulled out. Yet the media very quickly stopped telling the story as the promoter’s crisis. This is a case about the difference between formal authorship of a decision and reputational responsibility for its consequences.