Lyptus Research
Focus area
Dangerous Capability Evaluations
We measure the dangerous capabilities of frontier models and how quickly they are advancing. Our current focus is biosecurity and offensive cyber.
Read the offensive cyber studyLyptus Research is a not-for-profit AI safety research group based in Australia.
Frontier AI capabilities are advancing rapidly. Over 300 prominent figures, including 15 Nobel Prize and Turing Award recipients and 11 former heads of state and ministers, have publicly warned that this progress poses unprecedented risks. We take that risk seriously, and conduct research to help understand and prepare for it.
Focus Areas
Focus area
Alignment
We study how AI systems shape each other's values, particularly as frontier models increasingly drive the training of their successors. Our current work tests whether upstream models imprint preference structures on the models they help train, even under training for ostensibly orthogonal goals.
Read the latest work