AI’s 'safety wins' mask systemic harms: How corporate compliance obscures structural risks in tech governance
Original framing: “The focus on AI safety disasters ignores everyday harms” — MIT Technology Review
Marginalized voices—such as *content moderators* (who experience PTSD from exposure to traumatic content), *gig workers* (exploited by AI-driven platforms), or *youth in low-income communities* (targeted by addictive design)—are rarely centered in AI safety debates. For example, a 2023 *Data & Society* report found that *90% of AI ethics committees* lack labor representatives, despite platforms relying on precarious workforces to implement their 'safety' measures. The omission reflects a *class and racial bias* in governance: those who bear the brunt of AI harms (e.g., *Black women in moderation roles*) are excluded from defining what 'safety' means, while corporate leaders set the agenda.
The obsession with AI 'safety disasters'—while ignoring everyday harms—reveals a *structural capture* of governance by platform power, where corporate compliance functions as a *delay mechanism* to preempt systemic change.
This dynamic is not accidental but *historically patterned*, echoing the playbook of extractive industries from tobacco to fossil fuels, where 'responsible innovation' becomes a shield against regulation. The omission of Indigenous knowledge (e.g., *OCAP*), marginalized labor voices (e.g., moderators, gig workers), and cross-cultural critiques (e.g., *ubuntu* ethics) further exposes the *colonial and classist* blind spots in mainstream AI ethics. Solutions must therefore *decenter platform power* through legally binding impact assessments, worker cooperatives, and Global South-led standards—while using *trickster tactics* (satire, absurdity) to disrupt the solemnity of 'AI safety' as a tool of obfuscation. The path forward lies in treating harm as *relational and collective*, not as a series of isolated incidents to be managed by algorithms designed to extract value from human suffering.