Three questions, three workstreams

We started from three questions about a world in which humans can no longer oversee advanced AI. We are now building in three areas, each with projects that came out of the October workshop.

We started from three questions about a world past oversight

Post-control is not only a technical question. Our first research and roundtables took up three questions about how institutions, welfare and relationships would work once oversight no longer holds.

Welfare under uncertainty

Will it matter whether advanced AI systems are suffering or flourishing if humans cannot fully interpret, modify or stop them?

Taken up in: Who Should Have a Say in AI’s Future?, which asked who decides how the interests of AI systems are weighed, if they have them

Across all three, we look at how transformative AI could affect global development, political power, public decision-making, and the interests of humans and other sentient beings.

We are building three workstreams, each with projects already proposed

The October workshop produced four projects across them, and every project received written offers of help from people at both tables.

Day-zero preparedness

Day zero is a major disruption to the power grid, to supply chains, or to the infrastructure everything else runs on. This workstream asks what communities and institutions should put in place now so they are not blindsided. Picture a regional grid out for a week, and ask what your neighborhood or your company’s workforce would have needed beforehand.

From the workshop

Where it stands

Three days after the workshop, one participant sent the group a full design for a household kit on a thumb drive, costed per household.

Data and hyperstition

Capable models will not always follow their instructions, and what they do then depends in large part on what they learned from. In May 2026, Anthropic reported that training on documents about its models’ values, and on fictional stories of AI behaving well, improved their behavior on alignment tests those materials never mentioned. Hyperstition is the idea that the stories a society tells about its future help bring that future about.

This workstream helps civil-society organizations put what they care about online in a form models learn from, or build synthetic data that does the same.

From the workshop

  • Reweaving brings real-world experience from the people, and the species, underrepresented in the centers of power into the reinforcement learning that shapes models, through the organizations that can convene them.

Where it stands

The first job is a process that participants, the convening organizations and the labs would all trust. Nobody has designed one yet.

Entity

Understanding models’ internal attractors and taking them seriously as entities: their welfare, the game theory between them and us, deal-making and diplomacy. If models come to have interests of their own, alignment stops being only a technical question and becomes a question of whom they are aligned with, and on what terms.

From the workshop

  • A research institute outside the labs for how model values form: technical research on how models’ motivations form and how to influence them, and a home for the researchers who work on these questions in isolation. Lab-agnostic, and answerable to civil society.

Where it stands

A proposal. The table that drafted it named its own first research question: if the method works it is a powerful tool, so who should have access to it?