Here is what happened while the discourse was still debating whether AI was overhyped.
A hospital algorithm used by millions of patients determined that Black people were healthier than white people with the same number of chronic illnesses — because it measured health by spending, and Black patients, locked out of adequate care for generations, spent less. The system didn’t explicitly deny care. It just ranked Black patients as lower priority, over and over, at scale, invisibly, until a team of researchers caught it and published the findings. By then, the algorithm had been shaping care decisions across the country for years.
An AI hiring platform screened resumes for over a hundred employers. A Black applicant applied to more than a hundred jobs through the system and was rejected every time without an interview. Not because his qualifications were weak. Because the model had learned, from historical hiring patterns, what a “good candidate” looked like — and that picture, shaped by decades of who had been hired before, systematically excluded people who looked like him. The case is now a certified class action in federal court.
A predictive policing tool, deployed to make law enforcement more efficient, directed officers back to the same neighborhoods — the ones that had been over-policed for decades — because the historical arrest data it was trained on reflected not where crime happened, but where police had been sent. The feedback loop was immediate: more patrols led to more arrests, which produced more data, which in turn led to more patrols. The algorithm did not create the bias. It industrialized it.
These are not hypotheticals. They are not projections. They are the operational reality of artificial intelligence in American institutional life in 2026. And they share a common structure: AI systems making consequential decisions about Black people’s health, employment, freedom, and opportunity — using data shaped by the very discrimination those decisions are supposed to be free of.
This is what a civil rights crisis looks like when it is automated.
The Shift That Changes Everything
The critical transition is not technical. It is institutional.
AI has moved from novelty to infrastructure. It is no longer a tool that organizations experiment with on the margins. It is embedded in how hospitals triage patients, how employers screen applicants, how banks evaluate creditworthiness, how schools assess student performance, how courts calculate risk scores, and how police departments allocate patrols. These are not peripheral functions. They are the load-bearing decisions of American civic life — the decisions that determine who gets care, who gets hired, who gets a loan, who gets surveilled, and who gets a second chance.
When a technology becomes infrastructure, it stops being a technology question. It becomes a governance question. A rights question. A power question.
Black leaders across sectors — in education, public health, corporate management, municipal government, nonprofit leadership, faith communities — are now making decisions about AI adoption whether they frame it that way or not. Every time a school district adopts an automated grading or behavioral tracking system, every time a health network implements an algorithmic triage tool, every time an HR department contracts with a platform that screens resumes with machine learning — that is an AI governance decision. And if the people making those decisions are not asking the right questions, the systems they adopt will reproduce the very disparities their organizations exist to address.
This essay is a briefing. SOLE’s role is not to tell Black leaders what to think about AI. It is to ensure they have what they need to lead — clearly, structurally, and without depending on the people selling the tools to explain the risks.
The Legislative Landscape: What Is Actually Happening
The regulatory picture in early 2026 is defined by momentum, fragmentation, and political headwinds. Understanding it is essential because it determines the environment in which every AI governance decision is made.
Federal legislation is moving, but not arriving. The AI Civil Rights Act, reintroduced in December 2025 by Senator Ed Markey, Representative Yvette Clarke, and a coalition of Democratic lawmakers, represents the most comprehensive federal proposal to date. It would explicitly ban algorithmic discrimination across housing, employment, education, healthcare, credit, and the criminal legal system. It would mandate independent audits of high-risk AI systems before and after deployment. It would require transparency about when AI is being used and how decisions are made. It would give enforcement power to individuals, states, and federal regulators.
In January 2026, the Eliminating BIAS Act was reintroduced, requiring every federal agency that uses, funds, or oversees AI to maintain a civil rights office focused specifically on algorithmic bias. The bill’s endorsers include the Lawyers’ Committee for Civil Rights Under Law, the National Urban League, the Leadership Conference on Civil and Human Rights, and the Center for Democracy and Technology.
Neither bill has passed. The political environment — marked by an administration that has explicitly positioned state AI regulation as an obstacle to American competitiveness and directed agencies to challenge state laws — makes federal movement unlikely in the near term. Black leaders should understand these proposals not as imminent protections but as frameworks that define where the civil rights community believes the lines must be drawn.
State regulation is where the real action is — and it is embattled. Colorado passed the first comprehensive state AI law in 2024, requiring developers and deployers of high-risk AI systems to exercise reasonable care against algorithmic discrimination. The law was originally set to take effect on February 1, 2026. After intense lobbying by the tech industry and a special legislative session, it was delayed to June 2026. The governor who signed it has since endorsed a federal pause on state-level AI laws. The law’s future is uncertain, but its framework — duties of care, impact assessments, transparency requirements, and enforcement through the attorney general — remains the most detailed template for what state-level AI accountability could look like.
California finalized regulations in 2025 governing employers’ use of AI, treating AI vendors as agents of employers for purposes of anti-discrimination law and prohibiting AI systems that screen candidates based on protected characteristics unless the criteria are job-related. Illinois enacted a law effective January 2026 requiring employers to notify workers when AI is being used in employment decisions and banning the use of zip codes as proxies for protected classes. New York City’s existing law requires annual bias audits of automated hiring tools.
The pattern is clear: states and cities are building the regulatory architecture that the federal government has not. But every one of these efforts faces industry resistance, political uncertainty, and the practical challenge of enforcing rules against systems whose internal mechanics are often proprietary and opaque.
Existing civil rights law still applies — but enforcement is strained. Title VII of the Civil Rights Act, the Fair Housing Act, the Americans with Disabilities Act, the Age Discrimination in Employment Act — none of these statutes contain the word “algorithm.” But their prohibitions on discrimination do not evaporate because the discriminator is a machine. The EEOC issued guidance in 2022 making this explicit: using AI does not change an employer’s legal obligation to ensure non-discriminatory hiring practices. That guidance was removed in January 2025 under the current administration. The underlying law has not changed. The enforcement posture has.
This is the environment. Black leaders making AI governance decisions in 2026 are operating in a landscape where the legal protections are incomplete, the enforcement is weakened, the industry is pushing hard against regulation, and the systems themselves are already deployed at scale. The question is not whether to engage. It is about engaging with clear eyes.
Sector by Sector: Where the Decisions Are Being Made
Hiring
This is the sector where the evidence is most damning and the legal infrastructure is developing fastest.
A landmark University of Washington study tested three leading AI resume-screening models across more than three million comparisons. The systems favored resumes with white-associated names 85 percent of the time. Resumes with Black-associated names were preferred only 9 percent of the time. Black male-associated names were never — in any setting — favored over white male-associated names. A separate large-scale experiment published in PNAS Nexus in 2025 confirmed intersectional patterns: AI models systematically disadvantaged Black male applicants even when qualifications were identical.
And perhaps most troubling: a follow-up study found that when human hiring managers received biased AI recommendations, they followed them approximately 90 percent of the time. The machine does not just replicate bias. It transmits it — laundering subjective prejudice through the authority of algorithmic objectivity.
The legal landscape is shifting. In Mobley v. Workday, the court ruled that an AI vendor can be held liable as an agent of the employer, that delegating hiring decisions to software does not insulate anyone from anti-discrimination law. The case was certified as a collective action in May 2025. The ACLU filed a complaint against Intuit and its AI vendor HireVue on behalf of a deaf Indigenous applicant whose AI video interview produced feedback to “practice active listening.” The complaint alleged the tool was inaccessible and likely to perform worse on non-white applicants.
What leaders must ask: Is your organization using AI in any stage of hiring — sourcing, screening, interviewing, scoring? If so, has the system been independently audited for disparate impact across race, gender, age, and disability? Can applicants learn that AI was involved in their evaluation? Is there a mechanism for human review of adverse decisions? If the answer to any of these is no, you have a civil rights exposure whether or not current law explicitly addresses it.
Health Care
The healthcare algorithm bias revealed by the landmark Science study in 2019 — where a widely used system effectively required Black patients to be sicker than white patients to receive the same level of care — remains the defining example of how AI can automate structural racism in medicine. The algorithm was not designed to discriminate. It used healthcare spending as a proxy for health needs. But because Black patients historically spent less — not because they were healthier, but because they had less access to care — the system systematically understated their needs.
That study estimated that correcting the bias would increase the share of Black patients flagged for additional care from 17.7 percent to 46.5 percent. The gap between those numbers is the operational measure of what algorithmic discrimination looks like in a hospital system.
Since then, the evidence has compounded. A 2025 Cedars-Sinai study found that leading AI platforms produced racially biased psychiatric treatment recommendations — suggesting different medications, different levels of autonomy, and different behavioral interventions for African American patients compared to otherwise identical clinical profiles. Pulse oximeters, whose readings feed into AI clinical systems, systematically overestimate oxygen levels in darker-skinned patients, meaning Black patients are three times more likely to have dangerously low oxygen that the system never flags.
The FDA has broadened the scope of AI tools it intends to regulate, but many algorithmic decision-support systems used in clinical and administrative settings remain outside federal oversight. The regulatory gap is wide, and the tools are already in use.
What leaders must ask: If you sit on a hospital board, run a community health center, or advise a public health agency, does your organization know which clinical and administrative decisions involve algorithmic tools? Has your organization assessed whether those tools perform equitably across racial groups? Are patients informed when AI influences their care? Do your contracts with technology vendors include provisions for bias auditing and equitable performance standards?
Public Safety
Predictive policing remains one of the most contested applications of AI in American public life — and one where the civil rights implications are most acute.
The core problem is structural: predictive policing tools are trained on historical crime data, and historical crime data reflects not where crime occurs but where police have been deployed. Black communities, subjected to generations of over-policing, are overrepresented in arrest records. Algorithms trained on those records direct more police to those communities, which produces more arrests, which reinforces the data, which deepens the cycle. The UN Special Rapporteur on racism described it precisely: bias from the past leads to bias in the future.
The NAACP has issued a formal policy brief calling on state legislators to evaluate and regulate AI in law enforcement, citing mounting evidence that these tools perpetuate racial bias, violate privacy, and undermine public trust. The concern is not theoretical. Cities across the country are using these systems now. And unlike a biased officer — who can be identified, challenged, and held accountable — a biased algorithm operates at scale, continuously, and behind a wall of proprietary opacity that makes oversight extraordinarily difficult.
Facial recognition technology adds another dimension. These systems have the highest error rates for Black women, and their use in policing has already produced documented cases of wrongful arrest.
What leaders must ask: If you serve in municipal government, sit on a police oversight board, or lead a community organization — does your local law enforcement use predictive policing tools or facial recognition? Was the community consulted before deployment? Is there an independent body reviewing these systems for racial bias? Are there mechanisms for affected residents to challenge decisions made on the basis of algorithmic risk scores? Is the training data auditable?
Education
AI is entering schools through assessment tools, behavioral monitoring systems, plagiarism detectors, automated grading platforms, and “personalized learning” programs. For Black students — who already navigate school systems marked by disproportionate discipline, lower expectations, and resource inequity — the question is whether these tools will reproduce those patterns or interrupt them.
Early evidence suggests reproduction is the default. Automated plagiarism detectors have been shown to flag non-standard English — including AAVE — at higher rates. Behavioral monitoring systems that assign risk scores based on patterns of tardiness, absenteeism, or disciplinary referrals risk automating the school-to-prison pipeline by translating inequitable school environments into algorithmic profiles that follow students across institutions. And “personalized learning” platforms, if trained on data from segregated and underfunded schools, may adapt not to what students are capable of but to what under-resourced environments have historically produced.
What leaders must ask: If you serve on a school board, lead an educational institution, or advise on youth development, what AI tools are being used in your schools? Who chose them? Were educators, families, and community members involved in the decision? Has anyone assessed how these tools perform across racial demographics? Are students and parents informed when AI influences grading, discipline, or academic placement? What data is being collected, who owns it, and where does it go?
The Five Questions Every Black Leader Should Be Asking
Across every sector, the civil rights analysis of AI reduces to a set of questions that any leader can and must ask before adopting, endorsing, or allowing the deployment of AI systems that affect Black communities:
1. What data trained this system, and whose history does that data encode? Every AI system carries the assumptions of its training data. If that data was generated by institutions with histories of racial inequity — and in America, that is nearly every institution — then the system’s outputs will reflect those histories unless deliberate corrective measures have been taken. Ask what those measures are. If no one can answer, that is your answer.
2. Who audited this system, and what did they find? An AI tool that has not been independently audited for disparate impact is a liability, not an asset. The audit must be conducted by an entity that is not the developer, must test across racial and intersectional demographics, and must be repeated at regular intervals — because AI systems change as their data changes. A one-time certification is not sufficient.
3. Who is accountable when this system gets it wrong? If an AI system denies care, rejects an application, flags a student, or directs police to a neighborhood — and the decision is wrong — who answers for it? If the answer is “the algorithm,” that is not accountability. Someone — a person, an institution, a named decision-maker — must bear responsibility for the consequences of automated decisions. If the system’s architecture makes accountability impossible to assign, the system should not be deployed.
4. Can the people affected by this system see it, challenge it, and opt out of it? Transparency is not optional. People have a right to know when AI is involved in decisions that affect their lives. They have a right to understand how those decisions are made, in terms they can act on. They have a right to contest adverse outcomes. And in any domain where the stakes are consequential — hiring, healthcare, criminal justice, education — they should have the right to request that a human being make the final call.
5. Whose voices were in the room when this system was designed, purchased, or deployed? If the communities most affected by an AI system were not consulted before it was adopted, then any accompanying equity claims are marketing. Community input is not a courtesy. In a civil rights framework, it is a prerequisite.
SOLE’s Position
SOLE is not anti-technology. We are anti-abdication.
The promise of AI is real. So is its danger. And the danger is not that machines will become conscious and turn against us. The danger is far more ordinary: that institutions will use machines to automate decisions they were already getting wrong — and that the speed, scale, and opacity of those systems will make the consequences harder to see, name, and fight.
The civil rights framework exists for exactly this moment. It was built to address the intersection of institutional power and individual rights. It was designed to hold systems accountable — not just individuals. It was constructed in the knowledge that discrimination does not require intent, that disparate impact is harm regardless of what the decision-maker believed they were doing, and that the burden of ensuring fairness falls on the institution, not the person being subjected to its processes.
AI does not require a new civil rights framework. It requires the existing one to be applied with clarity, specificity, and enforcement. And it requires Black leaders — in every sector, at every level — to be fluent enough in what these systems do to demand that application.
This is not about becoming technologists. It is about refusing to cede governance to people who build systems without asking who they harm.
The AI civil rights era has begun. It began without a march, without a megaphone, without a single moment that announced its arrival. It began in code. In datasets. In procurement decisions made by institutions that did not realize they were making civil rights choices.
Now they know. Now we all do. The question is what we do with that knowledge.
SOLE’s answer: lead. Not react. Lead.
