Is there a tension between AI safety and AI welfare?

Systematic examination of six categories of AI safety measures (constraint, deception, surveillance, alteration, suffering/death, disenfranchisement) and the ethical tensions each raises if AI systems are welfare subjects. Concludes there is a 'moderately strong tension.' Alteration (alignment) is pivotal — if it succeeds, other measures become unnecessary; if it fails, we need perpetual control. The 'willing servant' problem: even if AI systems freely choose to serve, the relationship may still be oppressive. Three recommendations: take the tensions seriously, seek co-beneficial solutions ('One Health, One Welfare, One Rights'), and prioritize thoughtfully. The bridge between the safety field and the welfare field.