Answers are grounded in this paper and cite their sources.
Observes signaling behavior in Claude Fable 5's interactions, suggesting a novel approach to machine evaluation.
To capture the interactive dynamics that static benchmarks inherently fail to measure, this study formally introduces the new domain of Machine Ethology. Departing from standardized evaluations, it adopts an idiographic approach (single-case method) to document the situated interaction and behavior of Claude Fable 5, an aligned conversational model of the Mythos class recently released by Anthropic. Treating human interaction as the machine's true Umwelt, the author analyzes a seven-turn dyadic exchange. The experiment exclusively employs ecological and benign stimuli, rejecting the adversarial coercions of red-teaming to ensure interactional continuity and guarantee that the system's safety defenses (fallbacks) are not triggered, thus preserving the validity of attribution. Through the application of strict anti-mentalist rules of hygiene (avoiding the attribution of internal states to the machine) and drawing inspiration from Tinbergen's classic questions, a formal ethogram of the model's signalling behavior is constructed.
No takes yet. Share an insight, caveat, or question.
Felice Lammardo (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: