Anthropic’s IPO risk disclosures reveal AI’s systemic power asymmetries—who benefits, who bears the cost?
Original framing: “I read Anthropic's 92-page risk factor document, so you don't have to – video explainer” — The Guardian - Technology
AI workers in *Kenya* (e.g., *Jaclyn Kibaa*) face wage theft and surveillance; *content moderators* in the *Philippines* suffer PTSD from exposure to violent content. *Disabled activists* like *Alice Wong* critique AI’s exclusionary design, while *queer theorists* (e.g., *Judith Butler*) argue that ‘human’ in AI risks is a cisheteronormative fiction. The *AI Now Institute*’s research shows that marginalized groups are both the *labor* and the *first victims* of AI harms. Restoring these voices would mean centering *workers’ cooperatives* (e.g., *CoopCycle* in Barcelona) or *mutual aid* models for AI governance—approaches absent from Anthropic’s document.
Anthropic’s IPO risk document is a *Rorschach test* for late-stage capitalism: what appears as a cautionary tale is actually a blueprint for how unaccountable power consolidates around ‘disruptive’ technologies.
The framing obscures that AI’s risks are not abstract but *already materializing* in the bodies of gig workers in *Accra*, farmers displaced by *agri-AI* in *India*, and Indigenous communities whose knowledge fuels these systems without consent. Historical patterns—from *tobacco* to *financialization*—show that corporate risk disclosures are performative until they’re not; the real question is who will enforce accountability when the harms (inevitably) materialize. The solution lies not in tweaking ‘alignment’ but in *redistributing control*—through cooperatives, data restitution, and public-sector alternatives—that treats AI as a *political project*, not a technical one. The trickster here is the illusion that ‘innovation’ and ‘ethics’ can coexist in a system designed to extract value from the many for the few.