KVIG Informative

AI ethics is not a single job or a single philosophy. It is a set of practices for finding and reducing harm from automated systems, carried out by researchers, safety teams, auditors, regulators, and sometimes the engineers who build the products.

The field gained public attention through research that put numbers on a problem. In 2018, Joy Buolamwini of the MIT Media Lab and Timnit Gebru, then a Stanford graduate student, tested three commercial facial-analysis systems. The systems misclassified the gender of light-skinned men at error rates no worse than 0.8 percent, but misclassified darker-skinned women at rates of 20.8, 34.5, and 34.7 percent. For the darkest-skinned women in the data set, two systems erred about 46 percent of the time. IBM’s chief architect for Watson, Ruchir Puri, said at the time that the company had built a new model that was more balanced across the benchmark Buolamwini used.

The concerns the field covers include bias and fairness, transparency, privacy, accountability, and safety, meaning whether increasingly capable systems behave as their builders intend. The work happens in several places: research teams that publish studies, internal review groups, red teams that try to break systems before release, and independent evaluators such as METR, a nonprofit that tests AI models for outside clients.

The Frameworks Companies Follow In the United States, the main reference is the AI Risk Management Framework that the National Institute of Standards and Technology published in January 2023. It is voluntary, and it organizes risk work into four functions: Govern, which sets accountability and policy; Map, which identifies context and potential harms; Measure, which tests and monitors systems; and Manage, which acts on the results. NIST added a Generative AI Profile in July 2024. There is no certification attached to it, which means a company can say it follows the framework without any outside party verifying that claim.

The EU’s Binding Rules The European Union’s AI Act, which entered into force on Aug. 1, 2024, is binding law with obligations that scale by risk. Its prohibited practices began applying on Feb. 2, 2025, and its rules for general-purpose AI models on Aug. 2, 2025. A 2026 amendment known as the Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on July 27, 2026 and postponed the obligations for stand-alone high-risk systems, which cover areas such as hiring, credit, and education, from Aug. 2, 2026 to Dec. 2, 2027. Obligations for AI embedded in regulated products moved to Aug. 2, 2028. The prohibitions and most transparency duties were not delayed.

Nevada’s Version of the Question State government already uses AI. Nevada’s Department of Employment, Training and Rehabilitation began using a Google-run AI program to help decide unemployment appeals, with human verification, to clear a backlog. The agency said the tool would speed the process, while some lawmakers raised concerns about transparency and consent. In 2025, Gov. Joe Lombardo signed Assembly Bill 406 on June 5, banning AI from providing mental or behavioral health care services, with civil penalties of up to $15,000 per violation. A broader bill, Senate Bill 199, sponsored by Sen. Dina Neal, would have set guardrails on several uses of AI but failed on June 3, 2025. In September 2026, lawmakers proposed basic standards for how state and local agencies use AI, and Sen. Edgar Flores, who chairs the committee, said there are open questions about how the technology is being used.

These frameworks and laws depend on people inside organizations who test, question, and sometimes object. What happens when those people disagree with their employers is the subject of the rest of this series, which looks at documented departures, the legal protections for workers who speak up, and what remains unsettled.