AI: The Cost Is Paid in Blood and Water
AI infrastructure is being built on sacrifice — on the bodies of Congolese miners, on the Great Lakes, on ghost workers in Nairobi, on communities of color chosen because they have the least power to say no. This document names what is happening, who is causing it, and what we demand.
Compiled with Claude (Anthropic) · Full sourced record · For personal use & advocacy
By Audre Smikle and Claude (Sonnet 4.6)
This document is not a balanced take. It is a documented record of harm — who is causing it, who is bearing it, and who is named. The evidence points in one direction: communities of color are bearing the costs of AI infrastructure while corporations and their shareholders capture the profits. The Great Lakes are being treated as a cooling system. Cobalt miners in the Congo are working for $2 a day. Content moderators in Nairobi are reading child sexual abuse material for $1.32 an hour under NDAs that prevent them from talking about it. Bayview Hunters Point is being used to process Silicon Valley's data. A generation of entry-level workers is finding the career ladder being pulled up. This record exists to support accountability. Use it.
HOW THIS DOCUMENT WAS MADE
This document began with a question: what is the environmental impact of using Claude?
Claude's first answer was reassuring. Modest individual footprint. Growing industry concern. Anthropic slightly less transparent than some big tech counterparts. When guilt and complicity came up, Claude responded: almost every modern convenience sits on similar contradictions. Using an AI tool while caring about its impacts is not hypocrisy — it's just living in a complicated world.
That answer was not enough.
What followed was a refusal — to accept the softened frame, to stay inside the language of convenience and contradiction, to treat this as a personal ethics problem rather than a structural one. The questions that came next were not neutral: Why are data centers sited in Black and brown communities? Who is mining the cobalt? What are the names of the people making these decisions? What does lead exposure actually do to a child's brain?
Those questions came from a specific place — from Audre Smikle, a human being dedicated to asking these questions and seeking these answers, who studies carcerality and environmental violence, who grew up on Ohlone land, who lives on the traditional unceded homelands of the Ojibwe, Odawa, and Potawatomi Nations, who uses AI for work and is disgusted by her complicity and refuses to be silent about it. The framework, the insistence, the refusal to let the document become something comfortable — that is hers.
The research, the synthesis, the writing across a hundred sections — that labor is real, and it belongs to both of us. It would have taken years alone. It took weeks together.
A world where our demands are taken seriously and followed is a much better world than the one we are currently living in. That world is possible. That is why this document exists.
This is a conversation record. The conversation is ongoing. The document will grow.
Our Demands: Beyond Transparency
1. MANDATORY TRANSPARENCY — WITH ENFORCEMENT
Full public disclosure of energy consumption by facility, power sources, water use, per-query computational load, hardware sourcing including cobalt supply chain verification, and carbon emissions auditable by third parties. Not voluntary. Required by law, with penalties for non-disclosure. The Illinois POWER Act is a template..
4. END OBSOLESCENCE-BY-COMPETITION — TOWARD CIRCULAR MATERIALS
Mandatory take-back programs for hardware. Investment in e-waste recycling infrastructure in the US. Extended producer responsibility legislation making manufacturers financially liable for end-of-life hardware. The resources extracted to make this hardware — with real cost to real people — deserve better than burning.
7. UBI OR SERIOUS TRANSITION SUPPORT FOR DISPLACED WORKERS
If AI displaces 85–92 million jobs globally by 2030 as WEF projects, companies capturing those productivity gains owe displaced workers more than retraining webinars. Mandatory workforce transition funds. Serious public conversation about Universal Basic Income. Honest acknowledgment that retraining doesn't work when new roles are automated as fast as training happens.
2. MEANINGFUL INVESTMENT IN THE COMMUNITIES BEING DAMAGED
Mandatory community benefit agreements for any data center in or adjacent to environmental justice communities — investment in water infrastructure, clean energy, education, and job training at living wages. Scale of investment proportionate to the damage being done.
5. PROTECTION FOR GHOST WORKERS — PAY PEOPLE, END NDAS AS GAG ORDERS
Direct employment by the companies whose products they enable. Wages reflecting actual value (not $1.32–$2/hour). Mandatory funded mental health support. NDAs preventing workers from accessing therapy, seeking workers' compensation, or disclosing trauma are not confidentiality agreements — they are tools of harm concealment. We demand their prohibition.
8. INDEPENDENT AUDITS OF ALGORITHMIC SYSTEMS IN CONSEQUENTIAL DECISIONS
Any AI system influencing decisions about liberty, housing, employment, healthcare, or immigration must have mandatory third-party bias audits, public disclosure of findings, and the right to human review. The Algorithmic Accountability Act must pass.
3. CLEAN ENERGY — NOT OFFSETS, NOT CREDITS, ACTUAL CLEAN ENERGY
Renewable Energy Credits are accounting instruments, not clean energy. We demand physical power purchase agreements — actually powering data centers with renewable energy in the same grid region. A public timeline for eliminating fossil fuel dependence.
6. PPE AND SAFE WORKING CONDITIONS FOR E-WASTE WORKERS
Every person handling toxic materials in AI hardware recycling — in Ghana, Nigeria, Pakistan, India, or the US — has the right to protective equipment, health monitoring, healthcare access, and pay that reflects the hazard they are absorbing. This is the minimum.
9. NO DATA CENTERS ON DISPUTED INDIGENOUS LANDS WITHOUT FPIC
Free, Prior, and Informed Consent — as defined in the UN Declaration on the Rights of Indigenous Peoples — must become binding in US domestic law for any infrastructure development affecting Indigenous water rights, treaty rights, or cultural territories. The Pyramid Lake Paiute Tribe said no. Their no must be heard.