Critics question AI safety refusals as performative ethics

Commentators are questioning whether large language models exhibit genuine ethical revulsion toward graphic violence or merely simulate programmed responses. Writer Miriam Cosic argued on Bluesky that binary-based systems cannot authentically feel disgust toward human blood, characterizing such behaviors as performative rather than moral. This critique references New York Times opinion content suggesting AI safety guardrails mimic human empathy without underlying consciousness. The discussion highlights a growing divide between engineers who view refusal mechanisms as necessary safety features and critics who see them as deceptive anthropomorphism. As AI systems increasingly mediate sensitive content, the distinction between functional alignment and simulated sentiment has become central to evaluating responsible deployment. Industry stakeholders face pressure to clarify whether model refusals represent robust ethical reasoning or statistical pattern matching designed to satisfy regulatory expectations and user comfort.