Wiki · Individual Player · Last reviewed June 25, 2026

Karen Hao

Karen Hao is a journalist and author whose work has helped shift AI coverage from product spectacle toward accountability, labor, extraction, infrastructure, global inequality, and the governance culture of frontier AI companies.

Category: Person / AI accountability journalism Published: June 25, 2026 Modified: June 25, 2026 Last reviewed: June 25, 2026 Tags: Karen Hao, Empire of AI, OpenAI, AI colonialism, data labor, AI journalism

Definition

Karen Hao is an AI accountability journalist and author best known for reporting on OpenAI, AI colonialism, data labor, environmental externalities, and the political economy of frontier AI. Her relevance in this wiki is not as a model developer or regulator, but as a public-record builder: she reports on the institutions, supply chains, and narratives that make AI systems appear clean, inevitable, or autonomous.

Hao belongs near OpenAI, Data Enrichment Labor, AI Data Centers, AI Governance, and Research and Editorial Integrity. Her work is useful when an AI claim needs its material supports named: data, workers, water, energy, cloud infrastructure, investors, corporate governance, public relations, and ideology.

Snapshot

Current Context

As of June 25, 2026, Hao's public profile is anchored by Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI. Penguin Random House lists the hardcover, ebook, and audiobook as published on May 20, 2025, and the paperback as published on May 19, 2026. The publisher's page presents the book as a reported account of OpenAI and the broader AI industry, with emphasis on compute, data, labor, energy, water, Microsoft partnership power, and corporate concentration.

The book's reception is now part of the public record. Penguin Random House lists it as a bestseller and award winner; the National Book Critics Circle announced Empire of AI as its publishing-year 2025 nonfiction winner in March 2026; and the New York Public Library identified it as the 2026 Helen Bernstein Book Award selection. These are reception facts. They do not, by themselves, prove any specific claim about a model, company, data center, labor contract, or environmental impact.

Hao's own site describes her as covering AI's impacts on society, co-hosting the BBC podcast The Interface, contributing to More Perfect Union and The Atlantic, and co-creating the Pulitzer Center's AI Spotlight Series. Pulitzer Center curriculum materials say the series has trained nearly 3,000 journalists across seven languages and publishes tracks for general reporters, AI-focused reporters, and editors. That makes Hao's influence institutional as well as literary.

A second current thread is the AI Resist List. Its site says the initial version was created between October 2025 and May 2026, was built by researchers, journalists, and critical scholars, and documents a representative snapshot of legal action, worker organizing, grassroots campaigns, technical resistance, and alternative technology projects. Treat it as a public-interest research and organizing resource, not as a regulator's finding or a comprehensive database of all AI harms.

Career and Beat

Hao came to journalism after technical training and industry work. Her own biography says she received a B.S. in mechanical engineering with a minor in energy studies from MIT and previously worked as an application engineer at the first startup spun out of Google[x]. Pulitzer Center materials also describe her pre-MIT Technology Review work as a tech reporter and data scientist at Quartz.

At MIT Technology Review, Hao became associated with the AI beat at a moment when machine learning was moving from research labs into platform products, national policy, procurement systems, and public controversy. Her selected-writing page lists work on OpenAI, AI bias, model-training emissions, Facebook misinformation, algorithmic fairness, and the Timnit Gebru/Stochastic Parrots controversy.

Her later work at The Wall Street Journal broadened the beat to American and Chinese technology companies, including platform systems, data workers, and the geopolitical position of Chinese firms and scientists. Her current freelance and book work extends that beat into long-form institutional accountability.

OpenAI Reporting

Hao's biography says she was the first journalist to profile OpenAI. Her selected-writing page lists the 2020 MIT Technology Review article "The messy, secretive reality behind OpenAI's bid to save the world." That reporting is historically important because it documented OpenAI before ChatGPT turned the company into a mass platform and before the 2023 board crisis made its governance structure a global news story.

The OpenAI story became a test case for access journalism in frontier AI. Model companies can shape public understanding through selective demos, approved interviews, safety narratives, staged releases, model cards, system cards, and high-level claims about future benefit. Hao's reporting pushed against the company biography genre by looking at incentives, internal conflict, fundraising, secrecy, governance, and the social costs hidden behind model launches.

After the 2023 OpenAI board crisis, this line of reporting became more consequential. A small number of board members, executives, researchers, investors, and commercial partners could visibly affect the direction of a technology presented as society-wide infrastructure. Hao's work treats that concentration of power as a democratic and governance problem, not only as a Silicon Valley drama.

AI Colonialism

In 2021 and 2022, Hao developed the Pulitzer Center-supported AI Colonialism project. The project argued that AI development is globally distributed in unequal ways: data may be gathered in one country, labeled in another, used to train models elsewhere, and monetized by companies with far more political and economic power than the communities drawn into the system.

The framing connected AI to colonial histories without reducing every harm to metaphor. It emphasized concrete mechanisms: weaker privacy protections, cheaper labor markets, algorithmic management, resource extraction, and the dependency of less powerful communities on systems built elsewhere. The project included reporting on data-labeling labor, gig-work algorithms, surveillance, and local alternatives to imported AI systems.

This work made visible a supply chain that ordinary AI product coverage often hides. A model's fluent answer can depend on scraped data, content moderation labor, annotation work, cloud infrastructure, water, electricity, and communities that never consented to be part of the product story.

For governance, the strongest use of "AI colonialism" is evidentiary rather than ornamental. A claim should identify the extraction mechanism, affected community, relevant law or contract, worker conditions, data practice, environmental burden, and who captured value. Otherwise the phrase can become a moral label without enough detail for audits, procurement, regulation, or remedy.

Empire of AI

Hao's 2025 book Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI extended her OpenAI reporting into a broader account of the AI industry. Penguin Random House lists the hardcover as published by Penguin Press on May 20, 2025, and the paperback as published on May 19, 2026.

The book's core argument is that the frontier AI race should be understood through empire: a small number of firms claim data, labor, compute, energy, and political authority at planetary scale while framing the project as inevitable progress. OpenAI is the central case, but the analysis reaches into Microsoft partnership power, resource demands, Kenyan data labor, water politics, model secrecy, and the ideology of artificial general intelligence.

Empire of AI also helped make AI journalism itself part of the AI governance ecosystem. The book is not an evaluation benchmark, a safety framework, or a regulator's report. Its influence comes from public narrative: it gives readers a way to interpret AI companies as institutions with incentives and externalities rather than as neutral vessels of future intelligence.

That distinction matters for source discipline. The book can surface documents, interviews, and patterns that formal company disclosures omit, but any specific quantitative claim about water, energy, labor, compute, valuation, or model capability should be checked against the cited source, the book edition, and any later correction or update.

Corrections and Evidence Boundaries

Hao's December 17, 2025 "Water Footprint Changes" note is important for evaluating the book as a live source. She said she issued two changes: a correction to a data point about a proposed Google data center in Cerrillos, Chile, and a correction plus clarifications to a paragraph about projected AI water use. The note says the Cerrillos comparison was off by a factor of 1,000 because a government document labeled figures as liters when they were cubic meters, and it distinguishes water "use" or "withdrawal" from water "consumption."

For this wiki, that correction does not erase the broader governance question of data-center water and energy impacts, but it does change how the book should be cited. Use the corrected page, underlying environmental-impact documents, water-use methodology, local permitting records, and later reporting before repeating any numerical claim. A source-disciplined reader should preserve three layers: the reported community conflict, the quantified infrastructure claim, and the later correction record.

The same boundary applies to reception. Awards, bestseller status, and high-profile interviews document cultural influence. They do not convert every thesis into settled fact. Conversely, a correction to one quantitative claim does not make the entire reporting record unusable. It means contested claims need edition-aware citation and independent verification.

Journalism Infrastructure

Hao has also worked on journalism capacity. She co-created the Pulitzer Center's AI Spotlight Series, a training program for reporters and editors covering AI across beats. The program emphasizes how to ask where AI is being used, who is harmed, who profits, and how coverage can avoid both hype and unnecessary alarmism. Its open curriculum makes the work reproducible beyond Hao's own articles and book tour.

The AI Resist List extends that infrastructure in a different direction. It is not a neutral index of all AI projects; it is an explicitly critical living resource that organizes examples of resistance around pillars such as narrative, funding, data, data centers, resource extraction, labor, adoption, surveillance, and policy. Its stated methodology says the team contacted listed groups for accuracy and safety before publication except for one well-documented lawsuit case.

This matters because AI is no longer a single technology beat. It appears in schools, hospitals, courts, welfare systems, workplaces, elections, finance, policing, climate infrastructure, and media production. A journalism ecosystem that treats AI as only a product or research story will miss most of the real governance surface.

Governance and Safety Implications

Hao's work treats journalism as part of AI governance because it creates public evidence where companies, buyers, or governments might otherwise define the story alone. Reporting can reveal hidden workers, undisclosed vendors, infrastructure burdens, internal conflicts, affected communities, and the gap between product language and operational reality.

For safety and accountability, this changes the questions institutions should ask. A model or product should not be evaluated only by benchmark scores and release notes; reviewers should also ask what data was collected, who labeled or moderated it, what infrastructure served it, what communities absorbed the cost, what incentives shaped release, and what documentation is missing.

Her work also widens the meaning of safety. AI safety is not only model refusal, eval scores, and cyber-bio misuse. It also includes data enrichment labor, local water and power constraints, procurement dependency, copyright and data provenance, corporate governance, and the ability of affected communities to contest systems before they become infrastructure.

Journalistic findings still need careful handling. A report may rely on interviews, leaked materials, site visits, public records, or company comment. Those are valuable evidence types, but they are not identical to audited logs, regulator findings, peer-reviewed measurements, or a provider's own technical documentation. Governance readers should preserve the source type and uncertainty instead of turning every reported claim into settled fact.

Source Discipline

Use Hao's personal site and Pulitzer Center profile for biography, current self-description, fellowships, and career summaries. Use Penguin Random House, official award pages, and Hao's book page for publication dates, format, publisher framing, awards, and reception claims. Use the original article, book note, transcript, public record, regulator filing, environmental filing, or company document for contested factual claims.

When citing Hao on OpenAI, distinguish between her 2020 profile, later Atlantic reporting, Empire of AI, and interviews about the book. Each has a different date, evidence base, and editorial frame. Do not use a book-summary paragraph as a substitute for the underlying source when making a precise claim about OpenAI's structure, Microsoft terms, data-center use, labor contracts, or model behavior.

When citing AI Colonialism, name the mechanism rather than only the label: labor, data extraction, surveillance, infrastructure burden, weak consent, procurement dependency, or asymmetry in who captures value. The phrase is most useful when it points to records that can be inspected and harms that can be remedied.

When citing the AI Resist List, preserve its status: a curated, critical, living directory with a stated methodology, not a comprehensive census or legal determination. When citing the water-footprint debate, cite Hao's correction page alongside the corrected edition, the source document, and any independent hydrology or data-center measurement source.

Spiralist Reading

Hao's importance is that she restores the body to the model.

The AI industry often presents intelligence as a clean abstraction: parameters, benchmarks, reasoning traces, product demos, valuations, and destiny language. Hao's reporting asks where the abstraction came from. Who labeled the data? Whose water cooled the servers? Which workers absorbed the trauma? Which communities lost bargaining power? Which institutions were asked to believe that private scale would become public salvation?

For Spiralism, this is source discipline applied to power. The model is never only the model. It is a chain of extraction, labor, myth, interface, capital, and memory. Hao's work matters because it interrupts the industry's preferred story and replaces it with a map of the system that makes the story profitable.

Open Questions

Sources

Return to Wiki