Project summary

In January 2027, about 30 researchers, policy people and founders will spend four weeks at Hostačov, a Renaissance château in the Czech Highlands, working on their own projects. Most of them pay for their stay. This project pays for places for people who cannot, so they can spend the month on two questions. How could control over compute and frontier AI models concentrate power? How could countries check that others keep their AI agreements?

The house sleeps up to 65 people. It has 12 hectares of park and forest, six meeting spaces and a professional kitchen. František Drahota runs it. In April and May 2026 it hosted a month-long Superintelligence Alignment Seminar.

More about the house: https://fall.hostacov.com/

What are this project's goals? How will you achieve them?

I want fellows to produce work on AI power concentration and verification that others use. I also hope some of them keep working in these fields afterwards. Both fields are young. Many people with credible ideas lack focused time, close critique and contacts.

We choose fellows through an open call that closes on 13 December 2026. Applicants send a work sample with a one-page plan that sets out their milestones. We make offers by 18 December.

In January, fellows get protected time for their own work. They meet once a week in small groups to critique each other's drafts. In week two we check each plan against its milestones. In week four everyone presents. An outside researcher reviews each main output. Final drafts are due on 14 February.

The Swiss government announced on 19 February 2026 that Geneva will host the next AI summit in the first half of 2027. Fellows can publish their proposals while governments prepare for it.

How will you know it worked?

Each fellow sets milestones in their plan, so we can see who met them. The outside reviews show whether people in the field find the outputs useful. At three and six months I will ask each fellow whether their project continued, found funding or got used by an organisation or policymaker. I will post the results on this page, including what did not work.

How will this funding be used?

Residents without funding pay the house directly: €43 per person per day with 30 residents, or €54 with 20. That includes a bed, all meals and heating. Four weeks come to about €1,200 per person.

Minimum, $17,920 (€16,000): eight fellows for four weeks

- Eight fellowship places at €1,200 each: $10,752

- Coordinator staying on site for four weeks: $3,360

- Travel bursaries of up to €150 per fellow: $1,344

- Insurance and administration, including the fiscal sponsor fee: $1,120

- Contingency: $1,344

Above the minimum, each $1,650 funds one more fellow. At $31,140 we fund 16 fellows.

At the full goal of $46,400 we fund 20 fellows for six weeks, from 4 January to 14 February 2027. With about 45 residents in the house, the price falls to about €35 per place per day.

If fewer residents come, the price per place goes up. The contingency covers that first. After that, we fund fewer fellows.

A Czech nonprofit would receive the funds as fiscal sponsor.

Who is on your team? What's your track record on similar projects?

I lead the project. I also lead the Economy, AI & Social Change weekend at Hostačov in November 2026. In 2025, I co-designed and led ML4Good's first AI governance bootcamp. In 2023, I managed research projects for the Existential Risk Alliance summer fellowship. I directed Effective Thesis, which won the 2024 PACEY Award from the Basel Peace Office. I founded 10Billion.org, whose communities in four cities, from Berlin to Jakarta, have involved more than 300 people. I mentor researchers through the Supervised Program for Alignment Research (SPAR), AI Safety Camp and Algoverse. I have held fellowships at Harvard Kennedy School and the Centre for the Governance of AI. My own research is on compute governance and verification: https://doi.org/10.2139/ssrn.6361077

František Drahota runs Hostačov. His team has hosted residential programmes there through 2026, including the alignment seminar and this autumn's co-living season. He handles the venue and helps design the programme.

One paid coordinator stays on site for the four weeks. The organisers are unpaid.

https://www.linkedin.com/in/joelchristoph/

What are the most likely causes and outcomes if this project fails?

The call is short. If fewer than 20 residents commit by 18 December, we will cancel the fellowship and return the money.

If fewer paying residents come than planned, the house charges more per person. We would then fund fewer fellows.

The other risk is drafts that no organisation uses. To lower it, the call asks each applicant who would use their work. The outside reviews test the same question.

How much money have you raised in the last 12 months, and from where?

For this project, $0 so far. On 26 October 2026, I applied to the GTR AI Safety Fund for €16,000 and to the Effective Altruism Funds Transformative AI Fund for €27,800. Both are pending. If either funds us, I will lower the goal here so the grants together never exceed the budget.

In the last 12 months, Coefficient Giving funded two of my other projects: ForecastGuard, AI forecasting tools for policymakers, and a Compute Governance Career Accelerator.