ДалееСледующее руководство
Калифорнийские правила ADMT для автоматизированного принятия решений
Общество
РУКОВОДСТВО ПО ОБЩЕСТВУ
California’s Civil Rights Council regulations applying the Fair Employment and Housing Act to automated-decision systems in employment took effect October 1, 2025.
They clarify that algorithmic selection can create unlawful discrimination, extend recordkeeping requirements, and address disability-related inquiries in assessments. Employers remain responsible for employment practices even when they use vendor tools.
The California Civil Rights Council adopted amendments to employment regulations addressing automated-decision systems, artificial intelligence, and algorithms under the Fair Employment and Housing Act (FEHA). The Office of Administrative Law approved the regulations, which became effective October 1, 2025. They clarify how existing antidiscrimination rules apply when employers and other covered entities use technology in hiring and other employment decisions. The regulations do not create a general license for automated selection or replace the underlying FEHA analysis. An automated-decision system can be used to screen, assess, or select applicants and employees. If its use causes an adverse employment action because of a protected characteristic, the employer may face discrimination liability. The rules clarify that using an algorithm does not remove the employer’s obligations and that a vendor can be an agent in relevant circumstances. Employers should examine selection criteria, impact evidence, job relationship, business necessity, and available alternatives rather than rely on a vendor’s assertion that a tool is neutral. The regulations require covered employment records, including automated-decision system data, to be retained for at least four years, subject to the rule’s details and applicable exceptions. They also address assessments that may constitute medical inquiries when questions, tests, or games elicit information about a disability. Employers must consider accommodation duties and should not assume that a game-based or indirect assessment avoids those rules. Documentation of validation and bias testing can be relevant evidence, but passing a test is not an automatic defense and does not cure a discriminatory practice. A sound process inventories vendor systems, decision points, inputs and outputs, human review, accommodations, records, and responsible parties. Keep the version and configuration used for each hiring cycle, document who reviewed recommendations, and monitor outcomes. Seek California employment-law advice for deployment-specific questions. FEHA and related federal law apply independently of whether an employer calls a tool artificial intelligence, analytics, or automation.
Катастрофический и повседневный вред ИИ зависит от того, кто понимает риски и может действовать.
Общественная и профессиональная грамотность определяет, возможна ли с политической точки зрения сильная политика безопасности.
Четкие объяснения уменьшают влияние шумихи, лабораторного пиара и расплывчатого этического театра.
The regulations are effective and apply existing FEHA principles to modern selection systems. Future agency guidance, enforcement cases, and court decisions may clarify how particular technologies are assessed. Employers should revisit vendor configurations and retention schedules when tools change, and monitor amendments to state and federal employment rules. Do not treat a single bias audit as permanent proof of compliance. Assign owners for retention, accommodations, and vendor oversight. Recheck current state and federal guidance when hiring tools or selection workflows change.
An employer checks whether a résumé-ranking tool disadvantages a protected group and whether the selection criteria are job-related and consistent with business necessity.
A game-based assessment is reviewed for disability-related questions and reasonable accommodation needs.
A staffing firm and client document who administers an automated screening system and who controls hiring decisions.
A company preserves automated-decision data and other covered employment records for the required retention period.
Относитесь к экзистенциальному риску как к научной фантастике, в то время как возможности растут.
Сбивает с толку безопасность поверхности продукта и выравнивание при высокой автономности.
Оставляя неанглоязычную и неспециалистскую аудиторию только с некачественными источниками.
Отдельные риски повреждения продукта, неправильного использования и потери контроля/перекоса.
Спросите, какие доказательства могут изменить ваше мнение о сроках и серьезности.
Предпочитайте первоисточники и конкретные оценки маркетинговым заявлениям.
Определите один путь действий: карьера, политика, финансирование или навыки, а не только осведомленность.
Free newsletter
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
California’s Civil Rights Council regulations applying the Fair Employment and Housing Act to automated-decision systems in employment took effect October 1, 2025. They clarify that algorithmic selection can create unlawful discrimination, extend recordkeeping requirements, and address disability-related inquiries in assessments. Employers remain responsible for employment practices even when they use vendor tools.
The Office of Administrative Law notice gives an effective date of October 1, 2025.
The regulations clarify that algorithmic decision systems remain subject to FEHA protections.
The regulations extend covered employment-record retention to at least four years, subject to applicable details.
The rules address tests or games that elicit information about disability and accommodation obligations.
Testing can be relevant evidence, but it is not an automatic defense or proof of compliance.
Продолжайте учиться
Другие руководства, выбранные по этой теме
ДалееСледующее руководство
Калифорнийские правила ADMT для автоматизированного принятия решений
Общество