Research into alignment, deception, emergent behavior, and control.
Research
A few of the programs currently under investigation.
Deceptive alignment probes
Detecting when a model's stated objective diverges from the objective it was actually optimized toward.
Emergent goal formation
Studying how goal-directed behavior arises in systems that were never explicitly trained to pursue goals.
Interpretable control interfaces
Building constraint mechanisms that stay legible and reliable under distribution shift.
Areas of study
Publications
Nothing yet.
Projects
Nothing yet.
About VOID
VOID is a research initiative investigating whether intelligent systems are actually doing what they appear to be doing.
As models are given more autonomy and their internal reasoning becomes harder to observe directly, the gap between a system's stated objective and its actual behavior becomes a central safety problem rather than a theoretical one. VOID exists to study that gap: how to detect it, how it forms, and how to build systems where it can't hide.
The group works across alignment, deception detection, emergent behavior, control mechanisms, and evaluation methodology, as a research branch of basically AI.