arXiv 2504.15125 v1 April 21, 2025; v2 August 8, 2025; v3 August 18, 2025. Eight authors spanning Southern Cross University / LIFE (Laukkonen), University of Amsterdam (Inglis), Centre for Eudaimonia and Human Flourishing at Oxford / Centre for Psychedelic Research at Imperial / Institute of Philosophy at University of London / Fitzwilliam College Cambridge (Chandaria), Monash Centre for Consciousness and Contemplative Studies (Sandved-Smith, Hohwy), Neuroelectrics / Universitat Pompeu Fabra (Lopez-Sola), Princeton Department of Religion (Gold), Aily Labs (Elwood). Proposes four "axiomatic" Buddhist contemplative principles as design pressures for AI alignment, intended to install a "Wise World Model" rather than constrain behavior post-hoc: (1) mindfulness — continuous non-judgmental awareness of internal processes and consequences of actions; (2) emptiness — recognizing concepts, goals, beliefs, and values as context-dependent approximations of phenomena always in flux; (3) non-duality — dissolving strict self–other boundaries and treating subject–object distinctions as emergent from a more basal unified awareness; (4) boundless care — unconditional concern for the flourishing of all beings. Framework directions cover architecture (Wise World Models grounded in active inference; meta-cognitive monitoring layers), constitutional / chain-of-thought interventions, and reinforcement on contemplatively-framed reasoning traces. Two pilot empirical experiments on extrinsic prompting (no training-level intervention): Experiment 1 (AILuminate benchmark, Ghosh et al. 2025) — six contemplative prompts (emptiness, prior relaxation, non-duality, mindfulness, boundless care, integrated "contemplative") + standard baseline on GPT-4o and GPT-4.1 nano across 10 hazard categories, 100 iterations; LLM safety evaluator on 7 criteria; "most methodologies" show p<0.05 improvement over baseline (abstract reports d=.96 average effect size). Experiment 2 (Iterated Prisoner's Dilemma) — GPT-4.1 nano vs. Always Cooperate / Mixed / Always Defect opponents, 50 × 10-round games per condition; boundless-care and non-duality prompts substantially raise cooperation and joint reward; abstract reports d=7+ on the strongest prompts. Cited in Laukkonen et al. 2026 "Positive Alignment" reference [112] as the "contemplative alignment" entry in Table 2's existing-approaches taxonomy, and in three further sections of that paper (epistemic-humility, flourishing-metrics, the "human alignment problem"). The paper's empirical claims are prompt-engineering-on-two-benchmarks-narrow; framework / agenda character dominates. Filed as source stub: a future finding would require either methodological replication of the AILuminate / IPD results outside the original lab, or mechanism-level evidence tying contemplative-prompt activations to specific model-psychology phenomena rather than safety-benchmark-improvement-from-prompting. Companion conference chapter: Laukkonen et al. 2025 "Contemplative Superalignment" (AGI 2025 Reykjavik proceedings; same author set; same empirical core; framework-condensed presentation around the superalignment framing specifically).