We outline a research agenda for mutually beneficial artificial consciousness (MBAC): Artificial intelligence (AI) whose own subjective experience is positive and whose behaviour enhances human and non-human flourishing. Our starting point is the notion of beneficial states of consciousness (BSC)–qualitatively valued mind states such as kindness, joy, clarity, and non-duality–which we group into two practically useful categories: affective and contemplative. Acknowledging the diversity of biological and potential artificial minds, we argue that purposefully cultivating, modelling, and engineering BSC offers the most direct route to MBAC. We ground this claim in a case study of compassion that abstracts five interacting layers–neural (information routing), autonomic (mode switching), hormonal (broadcasting), developmental (calibration), and trainable (plasticity)–thereby revealing a hierarchical control motif (detect, appraise, switch mode, broadcast, recalibrate) that can inform AI design. Building on this template, we propose an interconnected research program with four looping components: cultivation of BSC in humans; collection of high-resolution neural, somatic, and cardio phenomenological data; modelling of the resulting multiscale dynamics; and translation into AI architectures. By integrating insights from neuroscience, physiology, developmental psychology, contemplative studies, and AI engineering, we aim to lay a conceptual and methodological foundation for conscious machines whose inner lives and outward impacts are aligned with the flourishing of all sentient beings.