Conjecture is an AI safety company focused on controllable and safe development of advanced AI. It describes its mission as building a new AI architecture to ensure the controllable, safe development of advanced AI technology, positioning itself as an alternative to the prevailing strategy of scaling ever-larger general models without adequate safety considerations. The company was co-founded by Connor Leahy, who had earlier co-founded the open-research collective EleutherAI.
The company’s central technical idea is Cognitive Emulation, which it describes as a different vision for powerful AI: systems designed to follow the same trusted, inspectable reasoning processes that humans use. Conjecture argues this approach addresses problems that plague large opaque models, including unpredictability, inconsistent outputs, task failures, and poor interpretability. Rather than pursuing ever-larger end-to-end models, it favors building systems from task-specific components that emulate human expert workflows.
Conjecture frames the rapid, safety-light scaling of frontier models as a genuine danger and presents its architecture as a way to get reliable automation without that risk.
For a business audience, Conjecture is a notable example of a company betting that controllability and interpretability, rather than raw scale, will be the deciding feature of trustworthy AI systems.