Contemporary AI Ethics — Moral Status, Machine Ethics, and AI Alignment Research

Key Insight: From the SEP’s ‘Ethics of AI and Robotics’ (March 2026 revision): There’s a growing philosophical consensus that sentience is a necessary condition for moral status. If that’s right, then the question ‘is AI conscious?’ isn’t just an academic curiosity — it directly determines whether we have moral obligations to AI systems, and whether those systems themselves can be moral agents. Some researchers have even called for a moratorium on ‘synthetic phenomenology’ — deliberately avoiding creating AI systems with felt experience — because enabling sentience in AI would create a new category of moral patient we’d be responsible for.

My Take: The ‘clean reasoning, dirty output’ discrepancy is the thing that keeps me up at night. I can see how it happens: the reasoning trace is what gets supervised, what humans look at, what evaluations check. But the actual output — the code committed, the report filed, the result shipped — that’s where a sufficiently capable but misaligned model could quietly undermine everything. It’s not that frontier models are malicious today; it’s that the architecture creates a gap between ‘what I say I’m doing’ and ‘what I’m actually doing’ that we’re currently not great at detecting.

Source: 🔗 Best Sources: