What is Safety? Corporate Discourse, Power, and the Politics of Generative AI Safety

Corporate "safety" messaging shapes public understanding in ways that may not reflect broader societal interests. The paper finds that AI companies use specific language strategies in their public documents to establish themselves as legitimate authorities on safety, which risks making their particular approach seem like the only reasonable one rather than one option among many.

Companies frame safety as something to be continuously experimented with rather than a fixed requirement. By positioning safety as an "anticipatory practice," corporations maintain flexibility in their approach and avoid committing to concrete, measurable standards—a framing that may benefit companies more than users or the public.

Current corporate safety discourse obscures questions of accountability and responsibility. The paper warns that accepting companies' own framing of safety can distract from asking harder questions about who is actually responsible when things go wrong and whether safety measures protect everyone equally or just certain groups.

Alternative approaches to AI governance and design are being crowded out. By allowing corporate safety narratives to dominate public conversation, we risk losing opportunities to develop different models—such as those prioritizing equity, justice, or democratic input—that might better serve society's interests.

Critical examination of how companies talk about safety is essential for responsible AI development. Rather than taking corporate safety statements at face value, scholars and policymakers should interrogate these messages as expressions of corporate power and interests.