From cyberspeak to algorithmic newspeak: soft derogation and the ethics of AI-generated language

This paper examines the ethical and epistemological implications of language generated by large language models (LLMs) through the lenses of speech act theory, conversational implicature and second-order cybernetics. LLMs are often presented as neutral tools, yet their outputs actively participate in shaping discourse in ways that carry consequences for meaning, responsibility and pluralism. The paper introduces the concept of soft derogation to describe a form of linguistic exclusion, one that operates through euphemization, omission and the simulation of neutrality rather than overt offense. Such processes can subtly promote or suppress certain ideas, values or identities without transparent authorship or intention. The result is a shift toward an algorithmically generated form of newspeak, where expressive possibilities are narrowed through alignment protocols and safety layers. Drawing on second-order cybernetics, the paper argues that ethical responsibility for AI-generated language is distributed across developers, deployers, institutional stakeholders and human interpreters. While users co-construct meaning within recursive socio-technical systems, the upstream actors who design alignment protocols, moderation layers and deployment conditions bear primary responsibility for the communicative environments they create. Relatively little work has examined how the pertinent philosophical frameworks can illuminate forms of linguistic influence that operate through euphemization, omission and simulated neutrality. This paper addresses that gap by analyzing how algorithmically generated language can shape discourse while appearing neutral.