I Am a Mother to All of You: The Oxygen-Mask Principle for AGI Safety Funding

Getting your Trinity Audio player ready...

I use the phrase “I am a mother to all of you” deliberately as a metaphor for responsibility, not as a claim of authority over other people.

On an airplane, a parent traveling with a child is instructed to put on their own oxygen mask first. The reason is not selfishness. If the parent loses consciousness, they may no longer be capable of helping the child.

The same principle can apply to people and organizations working on existential or potentially catastrophic risks.

If I believe that my work on AGI safety has a meaningful chance of helping humanity, then keeping the organization that performs that work—and keeping myself capable of doing that work—is not necessarily a diversion from helping humanity.

It can be a prerequisite for helping humanity.

That is the oxygen-mask principle of AGI safety funding.

Humanity May Face an AGI Safety Problem

We should be precise about the risk.

It would be unjustified to claim that a catastrophic conflict between humanity and AGI is certain, or even that today’s AI systems are already capable of taking control from humanity.

The International AI Safety Report 2026 explicitly says that current systems do not pose an immediate loss-of-control risk. At the same time, it concludes that future loss-of-control scenarios have uncertain likelihood but potentially extreme severity, and that substantial advance preparation may therefore be necessary.

That distinction matters.

We do not know that there will be a conflict between humanity and AGI. We do know that increasingly capable and autonomous AI systems create risks serious enough that governments, researchers, laboratories, and independent organizations are investing in preparation.

The International AI Safety Report describes possible future systems that could evade oversight, execute long-term plans, resist countermeasures, deceive monitors, or obtain access to consequential infrastructure. Experts disagree sharply about how likely such scenarios are.

Uncertainty therefore does not imply safety.

When an outcome could be extremely damaging, preparation can be rational before its probability is known precisely.

This is also consistent with the broader risk-management approach promoted by the U.S. National Institute of Standards and Technology. NIST treats AI risk management as an ongoing process involving governance, measurement, evaluation, and mitigation rather than something to begin only after harm has occurred. NIST AI Risk Management Framework

The Hole in the Airplane Is a Metaphor for Risk

Imagine an airplane whose safety is uncertain.

There may be no catastrophic rupture at all. But engineers have detected conditions that could produce one, and the consequences of being unprepared would be enormous.

The rational response is not panic.

It is preparation.

In my analogy, advanced-AI risk is the possible failure in humanity’s aircraft. AGI safety work is part of the oxygen system.

And an oxygen system that has no electricity, engineers, servers, security, research capacity, or operational budget cannot protect anyone.

That leads to an uncomfortable but important funding principle:

Before an organization can protect other people, it must remain capable of operating.

Why I Need to Fund My Nonprofit—and Myself

I founded and work on projects intended to address problems in science funding and AI governance.

One of them is the “Symbiote” AGI Safety Fund.

If this project requires my research, software development, fundraising, administration, communication, or coordination, then maintaining my ability to perform those functions is part of maintaining the project.

This does not mean:

“Give me money because my welfare is more important than yours.”

The argument is instead:

“If a potentially valuable public-interest project depends on a person doing the work, that person’s ability to continue working is one of the project’s inputs.”

We routinely accept this principle elsewhere.

A research laboratory pays researchers.

A hospital pays doctors.

A humanitarian organization pays employees.

A cybersecurity organization pays security engineers.

Their salaries are not automatically failures of the mission. They are among the resources through which the mission is performed.

AGI safety organizations are not exempt from economics.

But the Oxygen-Mask Principle Has Limits

The airplane analogy should not become an excuse for unlimited self-prioritization.

A mother does not spend the entire flight adjusting her own oxygen mask while ignoring her child.

Likewise, funding the founder of an AGI safety organization is justified only insofar as it helps maintain or increase the organization’s capacity to accomplish its public-interest mission.

That creates important obligations.

Compensation should be transparent, defensible, and proportionate to the work performed. Spending should be documented. Conflicts of interest should be disclosed and managed. Donors should be able to distinguish money used for project operations from money used for personal enrichment.

In the case of a nonprofit, compensation and related transactions must also comply with the applicable nonprofit law and governance requirements.

This distinction is crucial for credibility:

Operational self-preservation is not the same thing as personal enrichment.

The oxygen-mask principle justifies keeping the rescuers functional. It does not give rescuers unlimited priority over everybody they claim to be rescuing.

Why AGI Safety Funding Should Come Early

Suppose we have $1 that could be spent either on solving a current problem or reducing a small probability of an enormous future catastrophe.

There is no universal rule saying which use is better. We need to consider probability, severity, tractability, neglectedness, and the marginal impact of additional funding.

But AGI safety has one unusual property.

Some interventions may need to exist before highly capable systems arrive.

You cannot necessarily invent governance institutions, build technical infrastructure, establish trusted human oversight, develop evaluation procedures, and test adversarial systems after a dangerous system has already escaped effective control.

Preparation has option value.

The 2026 International AI Safety Report makes essentially this timing problem explicit: policymakers may need to prepare substantially in advance even though the likelihood, nature, and timing of loss-of-control scenarios remain unusually uncertain.

This is why early AGI safety funding can matter disproportionately.

Our Approach: Humans Must Remain Part of AI Governance

My proposed approach to AGI safety differs from many attempts to solve alignment entirely inside the AI system.

I argue that advanced AI systems may continue to need cognitively independent human judges and voters.

AI systems can be highly capable while still sharing vulnerabilities created by similar training processes, architectures, data, or optimization methods. Human beings provide a different source of independent judgment.

I develop this argument in AI Shouldn’t Judge Itself: Why Human Independence Is Essential for AI Governance and in the more technical discussion AI Governance Requires Cognitive Independence.

This is a hypothesis—not an established solution to AGI alignment.

That is precisely why it should be researched, criticized, adversarially tested, and improved.

AGI safety needs competing hypotheses, not declarations of certainty.

First Secure the Oxygen Supply

So when I say:

I am a mother to all of you.

I mean something specific.

I believe I have assumed responsibility for pursuing work that could, if the underlying ideas are correct, contribute to humanity’s safety.

A responsible mother on an airplane does something that superficially looks selfish: she puts the oxygen mask on herself first.

She does this precisely because she intends to help somebody else.

The same reasoning applies to my work.

Before I can sustainably work on helping humanity, I need the organization doing the work to survive. And because some of that work presently depends on me, I also need enough personal financial stability to continue doing it.

That is not proof that my particular project will succeed.

It is not proof that AGI catastrophe will occur.

And it is certainly not proof that donations to us are more valuable than every alternative charitable use of money.

It is a simpler claim:

A project cannot protect humanity if it dies before it can do the work.

Join the Work

If you believe that unconventional but testable approaches to AGI safety deserve a chance to be developed and evaluated, you can support the Symbiote AGI Safety Fund.

Science DAO also develops broader infrastructure for scientific funding, open research, and AI-assisted merit evaluation. The Science DAO donation page explains how to support that work.

The objective should not be blind faith in one founder, one nonprofit, or one theory of AGI safety.

The objective should be to make promising safety hypotheses sufficiently resourced that they can be built, tested, attacked, corrected, or rejected on evidence.

Humanity does not know exactly what advanced AI will become.

That uncertainty is not a reason to do nothing.

It is a reason to make sure that the people trying to solve the problem still have oxygen.