ASI escaping human control and pursuing ends misaligned with our flourishing is a central form of catastrophic, existential risk and it will not be mitigated by making chatbots slightly more honest. We want to develop alignment from a set of disparate methods and intuitions into a science capable of coherently and formally articulating risks, guarantees, predictions and fundamental properties about the systems it seeks to study. We confront our participants directly with the hard, general, philosophical problems, build paradigms rather than tools and find solutions to a failure mode that is not merely inconvenient but fatal. With theory firmly over empirics, we dedicate ourselves to principled, careful deliberation rather than trial and error.
Study, Explain, and Try to Solve
Superintelligence Alignment
The AFFINE Seminar is a month-long residency to work on ambitious holistic alignment theory through a mix of peer-tutoring and direct mentorship. We use carefully designed workshops and teaching experiments to induce productive states of exploration in our participants and unleash them on the hardest, least legible questions the field has to offer.
Next seminar: 4–30 January 2027. Applications close 22 November 2026.
From the first cohort
What participants say
AFFINE was the best month of intellectual exploration I have had the opportunity to engage in, ever. … At AFFINE I had time to wander towards and through interesting ideas.
Legitimately one of the best times of my life. Wonderful people. Fantastic support. Exactly what I needed right now to learn more and lean into helping humanity as much as I can in a sustainable and thoughtful way.
AFFINE is the best attempt I know of at toeing the line between enabling curiosity-driven exploration, and keeping people's focus on 'the hard parts of the alignment problem'.
Our mission
What no one else is doing
1-Month Seminar
4–30 January 2027. 35 participants. Free, with accommodation, catering and travel covered.
Year-Long Fellowship
Offered to the ~10 most promising participants. Selection is aimed at collaborative, not individual, excellence.
Mentorship
All of our mentors are available to all participants for 1on1 conversations about theory, meta skills and strategy.
Our method
Teaching differently
Instead of locking you into a topic or making you a research assistant to one of our mentors, participants are empowered to think clearly through hands-on pedagogic exercises and guided by their own curiosity. Promising researchers currently get into alignment through tightly scoped technical projects. Committing to an approach of this type actively discourages frame-finding work, since it does not promise predictable, legible progress on short time-scales. We want to turn that structure on its head. AFFINE participants wield responsibility for their own research direction as it is refined through immersive workshops and question driven prompting. Peer tutoring allows for the quick proliferation and refactoring of relevant concepts and mentoring aimed at conveying generators rather than facts enables our newcoming researchers to improve, reframe and un-ask any question they come across.
We filter for people who are actually good at freely doing interesting things with their mind, and try to empower them to think about whatever they deem interesting.
Our mentors
No need to choose
Participants get access to all mentors and are not restricted to a few depending on what they apply for.
Our mentors represent the frontier of conceptual AI alignment research – from foundational theory at MIRI to safety work at DeepMind, Anthropic and Astera Institute. All of them will be available to the participants for 1on1 conversations about theory, meta skills and strategy. Below: the mentors of the 2026 seminar. The 2027 mentors are listed on the program page.
Abram Demski
AFFINE · Ex-MIRI
Kaarel Hänni
Independent Researcher · Ex-Mila
Linda Linsefors
Independent Researcher · Founder of AI Safety Camp
Lucius Bushnaq
GoodfireAI
Richard Ngo
Independent Researcher · Ex-OpenAI · Ex-Google DeepMind
Justin Shovelain
Convergence Analysis · Ex-MIRI
Steven Kaas
AI Safety Info · Ex-MIRI
Felix Harder
Independent Researcher
Ihor Kendiukhov
Biodyn · Theomachia Labs
Jobst Heitzig
Potsdam Institute for Climate Impact Research
Who this is for
You don’t need an impressive degree
We don’t need you to have an impressive degree, what we want is a solid mathematical toolkit and philosophical vigour. We choose participants based on philosophical aptitude and a basic foothold in mathematical formalism, not on the degree to which they share our opinions.
If you work on these problems: visit, give a talk or better yet a workshop, and spend some time poking our participants with interesting questions.