Groundless Alignment × AI Safety Camp
We are opening applications to three tracks with the AI Safety Camp, Virtual Edition, 2026. Join us for a 3-month online research program from January to April 2026. Find out more about each track by clicking through on the cards below!
Join us in designing infrastructure that takes advantage of the sweet spot when AI is extremely cheap, fast, less error-prone, yet still only moderately intelligent. There is a lot to be experimented with, and practiced, around elements of engineering and culture that can enable wholesome integration of AI capabilties without disempowering or killing us.
We explore the claim that AI will continue to struggle to have integrity. Moreover, this “struggle” may persist (at least, for a little while, hence the playful number ‘2037’) without contradicting trends of increasing intelligence, efficiency, general agenticness, etc. Participants will (among other things) contribute towards making specific (and conditional) bets and forecasts, explainers, responses to and reviews of counter-arguments, collated evidence for and against.
MoSSAIC stands for Management of Substrate-Sensitive AI Capabilities. Here you work on clarifying limitations in mechanistic interpretability approaches that may show up as AI systems become more adaptive. Support the longer term goals of enabling substrate-sensitive safety approaches that are robust to self-modifying systems