Who Is Really Shaping the Future of AI?

If the people building AI believe it could threaten humanity, why are they competing so hard to make it more powerful?

I wanted to understand that. Dario Amodei and Sam Altman warn about catastrophic risks while leading companies that develop increasingly capable AI. Elon Musk backed a call to pause advanced AI training while building a competitor of his own. Yann LeCun challenges the argument that restricting access is the way to keep the technology safe.

You could dismiss the warnings as marketing or the continued development as hypocrisy. But that would let everyone off too easily. Some of these people sincerely believe AI could produce extraordinary benefits and extraordinary dangers. They also believe they have a role in determining which future we get.

The question is why the rest of us should accept their judgment, and who can challenge it when it matters.

That question led to our new Techstrong.ai special report, Building AI, Fearing AI: The movements, money and personal rivalries behind AI’s power networks.

Download Building AI, Fearing AI (PDF)

The Founders Are Only Part of the Story

The relationships alone make for a compelling story. Musk and Altman helped establish OpenAI. Amodei worked there before leaving to build Anthropic. Former colleagues became competitors, carrying disputes over trust, commercialization and control into institutions with ambitions that reach far beyond selling software.

These relationships matter when companies are asking us to trust their safety commitments. Cooperation requires some confidence that a competitor will disclose an inconvenient finding. Independent review requires agreement about who qualifies as independent. Personal history can affect both.

But following the personalities takes you only so far.

Behind the familiar names are philanthropic donors, career organizations, fellowships, policy researchers, evaluators and governance bodies. Their connections help explain which approaches receive sustained support, which people enter influential positions and whose assessment gets heard.

Some of that power operates through formal rights. Some operates through the ability to fund an institution for years. Neither requires a public confrontation between CEOs to be consequential.

Follow the Institutions

Consider two records examined in the report. Georgetown announced a $55 million Open Philanthropy grant when its Center for Security and Emerging Technology launched in 2019. CSET announced another $42 million in 2021. Its account also described former staff working in the National Security Council, the Office of Science and Technology Policy and Congress.

That does not prove a donor controlled government decisions. It does show how private funding can support an institution with a route into public policy.

Then there is Anthropic’s Long-Term Benefit Trust. In April 2026, the company announced a trust appointment that brought trust-appointed directors to a board majority. That was exercised appointment authority, not simply a philosophical preference.

A structure intended to protect the public interest may be valuable. It still deserves questions about who selects its members, whom they answer to and how their decisions can be challenged.

The report follows those kinds of connections through funding, talent development, evaluation and governance. A donor does not need to dictate a research conclusion to influence a field. Choosing which institution can hire staff and pursue a question for five years is consequential on its own.

We also distinguish the movements too often bundled together: Effective altruism, longtermism, rationalism, effective accelerationism and techno-optimism. AI safety is a field of work, not proof of membership in any one of them. Shared arguments do not establish a common chain of command.

What emerges is a picture of influence that cannot be understood from company announcements alone.

Why Buyers Should Care

If you are choosing AI for your business, this can sound distant from the work of evaluating models, negotiating contracts and managing deployments. It isn’t.

Your supplier’s approach to risk influences what it tests and what it releases. The scope of an outside evaluation determines how much reassurance you should take from it. Governance determines who can overrule an objection. None of those questions is answered by a benchmark score.

The report examines evaluators’ independence, including financial support, access to models and personal conflicts. It also looks at the reputational damage EA experienced after FTX and why rejecting a movement’s label does not settle the separate question of institutional relationships.

Thiel and Palantir complicate the picture further. Historical philanthropy and commercial partnerships cross boundaries that public arguments can make look absolute. We distinguish those relationships rather than treating every connection as evidence of ideological allegiance.

This is not a report claiming a secret society controls AI. The records do not establish that. They establish enough institutional influence to warrant scrutiny without that claim.

I am not arguing that we should abandon AI or dismiss genuine safety concerns. I am arguing that people claiming to protect humanity should expect questions about their authority, their funding and their accountability. So should the organizations supporting them.

The full report brings the records together, with dated funding disclosures, institutional connections and source notes. It asks readers to look beyond the founder onstage and examine who helps shape the decisions behind the product.

The closing question is the one I keep coming back to: Who can tell them no, and make it stick?

Read the full Techstrong.ai special report: Building AI, Fearing AI.