Rules tracker · dated and sourced
A list of the laws, regulations and official guidance that govern how AI is used to screen, rank and advise people looking for work. Each entry names its primary source.
About the tracker
What it is
One entry per rule: a law, regulation, piece of official guidance or standard that governs how automated systems screen, rank or advise job seekers. Each names its jurisdiction, where it stands, what it requires and the primary source it comes from.
Why the lab keeps it
HyperQuark Intelligence Labs studies how AI advises, judges and governs people looking for work. The rules are part of that picture. Before anyone can ask whether a hiring system behaves well, they need to know which rules apply, and where.
Who it is for
Employers and vendors who use AI to screen or rank candidates. Job seekers who want to know what protections exist where they apply. Researchers, journalists and policy staff who need a sourced place to start.
Anatomy of an entry
How to read an entry
Jurisdiction
The place whose rule it is: a country, a state or city within one, or a group of countries.
Instrument
The law, regulation, guidance or standard by name, with its citation and what kind of instrument it is.
Status
Where it stands legally (see the key), and the date it takes effect when one is set.
What it requires
The lab's plain summary of the obligation, and who it applies to. A starting point, not a substitute for the text.
Source
A link to the primary source, often with a short quote from it, so you can see the words the summary rests on.
Checked
The day the entry was last read against its source, who read it, and whether that reading is confirmed.
Status key
- In force
- Its obligations apply now.
- Partly in force
- Some provisions apply now; others start later.
- Enacted, not yet in force
- Passed or adopted, with obligations that start on a later date.
- Proposed
- A bill or draft. Not binding unless it is adopted.
- Withdrawn
- Proposed and then dropped. Kept for the record.
- Superseded
- Replaced by a later instrument. Kept for the record.
Checked key
- Confirmed
- The summary has been checked against the primary source.
- Unconfirmed
- Published before that check was complete, by a recorded decision. Read the source first.
Practical
How to use this
Narrow it
Filter by jurisdiction for the places where you hire or apply, and by status to separate what applies now from what is coming.
Go to the source
Read what an entry requires, then open its primary source. The summary is a starting point; the source is the rule.
Mind the date
Each entry says when it was last checked. A rule can change after that day, so treat the date as part of the entry.
Cite or reuse it
Each title links to that entry alone. The CSV holds every published entry, under CC BY 4.0: reuse it freely, crediting HyperQuark Intelligence Labs.
The tracker
The rules
- IndiaIn force
- What it requires
- Voluntary, non-binding. Lists bias and discrimination as an AI risk, giving employment decisions as an example. Recommends that anyone developing or deploying AI comply with existing law and adopt voluntary fairness, non-discrimination and transparency measures. It also recommends a grievance redressal mechanism for AI harms, transparency reports, and tools such as algorithmic auditing and automated bias detection.
- Applies to
- Any person developing or deploying AI systems in India (industry guidance), plus regulators and government bodies.
Bias and discrimination, such as the use of inaccurate data to make a decision about future employment, which may result in loss of opportunity or livelihood.
Primary source: static.pib.gov.in Confirmed
Notes
The 5 Nov 2025 release date is confirmed by PIB release 2186639 and by PIB 2206769 (19 Dec 2025, a Rajya Sabha reply). That reply says the Guidelines are principle-based and not prescriptive, add no new statutory mechanisms, and say no new horizontal AI law is needed for now. Enforcement is left to existing laws and sectoral regulators. The guidelines also mention 'transparency in automated decisions that have an impact on life or livelihood' as a use for techno-legal measures. Per PIB 2252739, MeitY set up the AI Governance and Economic Group on 16 Apr 2026 to act on these guidelines.
- IndiaIn force
- What it requires
- Voluntary. Sets procedures for testing AI and machine-learning systems for bias across protected attributes and producing fairness scores, either by the developer's own self-assessment or by an independent third-party audit. It gives job candidature as an example of a decision that gender should not influence.
- Applies to
- Organisations and individuals developing AI systems, and auditors; voluntary, not limited to telecom.
The feature that must not influence the decision process of a machine learning algorithm. For example, the gender of a person should not affect the job candidature in general.
Primary source: tec.gov.in Confirmed
Notes
The document says 'Release : July, 2023' but gives no exact day, so effective_date is null. It says 'the intention is not to make the framework a mandatory regulatory assessment tool'. The India AI Governance Guidelines (2025) name it as a voluntary standard for bias detection and mitigation.
- UAEIn force
- What it requires
- Data subjects may object to decisions resulting from automated processing, including profiling, especially those with legal or adverse effects, except where agreed by contract, required by other UAE law, or consented to in advance. Controllers must provide human review of such decisions on request, disclose them on request, and assess impact before systematic automated evaluation with legal or serious effects.
- Applies to
- Controllers and processors in the UAE or processing data of people there; excludes government entities and free-zone companies with their own data-protection laws.
the right to object to any decisions resulting from automated processing, including profiling, particularly those decisions which have legal impact on or adversely affect the Data Subject
Primary source: uaelegislation.gov.ae Confirmed
Notes
The official portal lists: Issued 20 Sep 2021, Effective 02 Jan 2022, Official Gazette 26 Sep 2021 (No. 712 supplement), status Active. Article 31 says the law comes into force on 02 January 2022. Article 28 required the Executive Regulations within six months of promulgation. On 2026-09-28 the law's related-legislation tab listed only the KYC Digital Platform decree-law (No. 30 of 2024) and the Child Digital Safety decree-law (No. 26 of 2025). Portal keyword searches for 'Protection of Personal Data' and 'Executive Regulation Personal Data' returned no Executive Regulations for this law. Article 29 gives controllers and processors six months from the date the Executive Regulations are issued to comply, so no compliance deadline is fixed yet. The English text is a translation and the Arabic prevails. Article 18 has no wording specific to hiring. Article 2(2) also excludes government data and entities, security and judicial authorities, health and banking data covered by their own laws, and free-zone companies with their own data-protection laws (such as DIFC and ADGM).
- UAEIn force
- What it requires
- Sets principles for developing and using AI. They include reducing algorithmic bias so there is no exclusion or discrimination, being transparent about how systems make decisions, keeping human oversight to correct errors or biases, protecting privacy, and complying with UAE law. It has no provisions specific to employment or hiring.
- Applies to
- Target audience listed as public sector, private companies, consumers, AI developers, researchers, educational institutions, research centres and the international community.
The Charter emphasizes the irreplaceable value of human judgment and human oversight over AI
Primary source: uaelegislation.gov.ae Confirmed
Notes
The official legislation portal lists this as a policy with an issue date of 10 Jun 2024; the effective_date given here is that issue date, because no separate effective date is stated. The lead entity field is blank. Neither the portal nor u.ae describes it as binding or sets penalties. No federal AI guidance specific to recruitment or employment was found.
- UAE: DIFCIn force
- What it requires
- Individuals may object to decisions based solely on automated processing, including profiling, with legal or other seriously impactful consequences, and require manual review. Contract and explicit-consent exceptions need safeguards including manual review; no exception applies to minors. Controllers must disclose automated decision-making with meaningful information about its logic, and assess impact before High Risk Processing.
- Applies to
- Controllers and processors incorporated in the DIFC, and others processing personal data in the DIFC as part of stable arrangements.
A Data Subject shall have the right to object to any decision based solely on automated Processing, including Profiling, which produces legal consequences concerning him
Primary source: assets.difc.com Confirmed
Notes
Article 4 of the law text says it came into force on 1 July 2020. The DIFC legal database page for the law gives Enactment Date 21.05.2020 and Commencement Date 01.06.2020, which conflicts with Article 4; the date in the law text is used here. The quote is from the July 2025 consolidated version, as amended by DIFC Law No. 2 of 2022 and DIFC Laws Amendment Law No. 1 of 2025. The 2025 Amendment Law was enacted 8 July 2025 and came into force on the 5th business day after. It amends Articles 6, 28, 46, 59, 61, 62 and 64, adds Article 64A, and changes Schedule 2 fines (up to USD 100,000 for Article 38). It does not change the text of Article 38. Whether the 2022 amendment changed Article 38 was not checked. Article 38(2)(b) also exempts decisions authorised by Applicable Law with suitable safeguards. Article 38(6) limits solely automated decisions on Special Categories data to explicit consent or Substantial Public Interest. High Risk Processing Activities include systematic and extensive automated evaluation on which decisions with legal or similarly significant effects are based. Article 38 has no wording specific to hiring.
- UAE: DIFCIn force
- What it requires
- Deployers or operators of AI Systems processing personal data via apps or websites must give clear notice at first use, describing purposes, outputs and design principles. On request they must evidence certification, human-intervention triggers for unfair or discriminatory impact, and a register noting any solely automated decision-making. High Risk Processing requires certified Systems and an Autonomous Systems Officer.
- Applies to
- Deployers, operators and providers of autonomous or semi-autonomous Systems processing personal data subject to the DIFC Data Protection Law.
seek human intervention when Processing of Personal Data by the System may result in an unfair or discriminatory impact on a Data Subject
Primary source: assets.difc.com Confirmed
Notes
The cover of Consolidated Version No. 2 reads 'In force on 1 September 2023'. The DIFC Regulation 10 page says the updated regulations were enacted on September 1, 2023, and DIFC announced the enactment on 7 September 2023. Law firms and vendors say full enforcement began on 1 January 2026, but no DIFC primary source stating that date was found. DIFC's Regulation 10 Accreditation and Certification Framework says the timeline for implementing the Framework is at the Commissioner's discretion. Regulation 10.3.3 bars High Risk Processing by a System until the Commissioner has set audit and certification requirements and the System meets them. DIFC's certification portal guide (DIFC-DP-GL-25 Rev. 01, updated 10 August 2026) describes a live certification process: DIFC fees of USD 100 or 200, a cap of USD 5,000 on Accredited Certification Body fees, and certificates valid for three years. The Regulation 10 page lists four Accredited Certification Bodies, one of which (Standard Chartered Bank) certifies only its own internal systems. Regulation 10.3.1 also requires Systems to be designed to be ethical, fair, transparent, secure and accountable. A Deployer is treated as the Controller. Amendments proposed in June 2026 are in a separate row. The regulation has no wording specific to hiring.
- UAE: DIFCProposed
- What it requires
- No new duties in force yet. DIFCA proposes to amend Regulation 10: adding 'Safety' as a design concept, setting privacy-by-design and default measures, using the broader term 'policy frameworks' in notices, and detailing Autonomous Systems Officer duties and skills. It also proposes a new Regulation 11 letting the Commissioner recognise accreditation and certification schemes. Comments closed 18 July 2026.
- Applies to
- Would apply to persons subject to the DIFC Data Protection Regulations, including deployers, operators and providers of AI Systems.
The Proposed Regulations are in draft form only. You should not act on them until they are formally enacted.
Primary source: assets.difc.com Confirmed
Notes
DIFC published the news release on 18 June 2026, with a 30-day consultation ending 18 July 2026. The paper says the regulations will come into force on a date that DIFCA will specify and publish after enactment. As of 2026-09-28, the DIFC legal database still links Consolidated Version No. 2 (in force 1 September 2023), and no enactment notice was found. The cover says Consultation Paper No. 3, but paragraph 1 calls it 'Consultation Paper No. 2 of 2026'.
- UAE: ADGMIn force
- What it requires
- Individuals have the right not to be subject to decisions based solely on automated processing, including profiling, with legal or similarly significant effects, unless needed for a contract, explicitly consented to, or authorised by law with notice. Contract and consent cases require human intervention and a chance to contest. Controllers must disclose such decision-making and meaningful information about its logic.
- Applies to
- Controllers and processors with an establishment in ADGM, for processing in the context of those activities, wherever the processing occurs.
The Data Subject has the right not to be subject to a decision based solely on automated Processing, including Profiling, which produces legal effects concerning him or her
Primary source: en.adgm.thomsonreuters.com Confirmed
Notes
The regulations were enacted on 11 February 2021 (date of enactment stated in the text). Section 63 brings them into force in two phases: 6 months after publication for establishments set up on or after publication, and 12 months after for existing ones. The ADGM Office of Data Protection webinar deck (4 August 2021) and ADGM's overview brochure, both on assets.adgm.com, give publication as 14 February 2021 and the effective dates as 14 August 2021 (entities established after 14 February 2021) and 14 February 2022 (entities established on or before that date). Under section 20(2)(c), the law-based exception applies only if the controller told the person in writing and the person did not ask for reconsideration within one month. Decisions under the exceptions must not use Special Categories data unless section 7(2)(a) or 7(2)(k) applies and safeguards are in place. Sections 11(2)(g), 12(2)(g) and 13(1)(h) require disclosure of automated decision-making and meaningful information about its logic. Section 34 requires a DPIA before processing likely to be high risk. The ADGM news pages on adgm.com were blocked by a bot check.
- UAE: ADGMIn force
- What it requires
- Explains Data Protection Regulations 2021 rule on solely automated decisions, citing a fully automated recruitment aptitude test as an example of automated processing. Says individuals must be able to express their view, ask for an explanation, challenge decisions, and object or demand human intervention. Human reviewers need meaningful authority to overturn system recommendations and must not routinely apply them.
- Applies to
- Controllers subject to the ADGM Data Protection Regulations 2021 that use automated decision-making or profiling.
A fully automated recruitment aptitude test which uses pre-programmed algorithms and criteria
Primary source: assets.adgm.com Confirmed
Notes
The brochure is dated May 2025 and gives no exact day or effective date. It does not say whether it is binding, and it does not cite a section number. It describes the right as 'essentially an obligation placed on Controllers' not to make such decisions. The same file is also listed at an adgm.com path under office-of-data-protection/resources/brochures-flyers. Separate ADGM DPIA guidance (Guidance on the DPR 2021, Part 4, VER1.0-11082021) lists using profiling or automated decision-making to help decide someone's access to a service, opportunity or benefit as processing that needs a DPIA.
- EUEnacted, not yet in force
- What it requires
- AI systems intended for recruiting or selecting people (placing targeted job ads, analysing and filtering job applications, evaluating candidates) are high-risk and must meet the Regulation's high-risk requirements. An Annex III system is not high-risk if it poses no significant risk (for example, only a narrow procedural or preparatory task), but one that profiles natural persons is always high-risk.
- Applies to
- Providers and deployers (for example employers or recruiters) of AI systems intended for recruitment or selection of natural persons.
intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates
Primary source: publications.europa.eu Confirmed
Notes
The Regulation entered into force on 1 August 2024, but this provision is not yet applicable. As originally published, Chapter III Sections 1-3 would have applied to Annex III systems from 2 August 2026. Regulation (EU) 2026/1744 moved that date to 2 December 2027 (Article 113, third paragraph, point (c) as amended). Article 6(4): a provider that considers an Annex III system not high-risk must document that assessment before placing the system on the market. Article 111(2) as amended: high-risk systems placed on the market or put into service before the Chapter III application date are covered only if their design changes significantly from that date. Commission draft guidelines on high-risk classification (Article 6(5)) were published 19 May 2026 and were still marked draft on 2026-09-28. Separately, Article 50(1) requires providers to design AI systems that interact directly with people so those people are told they are dealing with an AI system, unless that is obvious. It is in Chapter IV, which was not deferred, so it applies from 2 August 2026. EUR-Lex blocked automated fetching (HTTP 202, empty body), so the text was loaded from the Publications Office (same OJ text). ELI: http://data.europa.eu/eli/reg/2024/1689/oj.
- EUEnacted, not yet in force
- What it requires
- Deployers must use the system per its instructions, assign human oversight to competent, trained people with authority, keep input data they control relevant and sufficiently representative, monitor operation and report risks and serious incidents, keep automatically generated logs at least six months, inform workers' representatives and affected workers before workplace use, and inform people subject to the system.
- Applies to
- Deployers of Annex III high-risk AI systems, including recruitment and selection systems; paragraph 7 applies to deployers who are employers.
deployers of high-risk AI systems referred to in Annex III that make decisions or assist in making decisions related to natural persons shall inform the natural persons
Primary source: publications.europa.eu Confirmed
Notes
Article 26 is in Chapter III, Section 3. For Annex III systems it applies from 2 December 2027 under Article 113 as amended by Regulation (EU) 2026/1744 (originally 2 August 2026). Regulation 2026/1744 did not amend Article 26. Article 26(9): deployers use the provider's Article 13 information for their GDPR Article 35 data protection impact assessment. By its text, the Article 27 fundamental rights impact assessment applies to public-law bodies, private entities providing public services, and deployers of Annex III point 5(b)/(c) systems; Regulation 2026/1744 amended Article 27(4)-(5). The text does not say whether 'affected workers' in Article 26(7) includes job applicants. The log-keeping duty in Article 26(6) covers logs to the extent they are under the deployer's control. EUR-Lex blocked automated fetching, so the text was loaded from the Publications Office.
- EUIn force
- What it requires
- Placing on the market, putting into service for this purpose, or using AI systems to infer a person's emotions in the workplace is prohibited, except for medical or safety reasons. The Commission's non-binding guidelines state that 'workplace' covers candidates during selection and hiring, and that using emotion recognition during recruitment is prohibited.
- Applies to
- Anyone placing on the EU market, putting into service or using AI systems that infer emotions in the workplace, including in recruitment.
the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions
Primary source: publications.europa.eu Confirmed
Notes
Chapters I and II apply from 2 February 2025 (Article 113, point (a)). Regulation (EU) 2026/1744 added new Article 5 prohibitions (points (ba) and (bb), applying from 2 December 2026) but did not change point (f). Article 3(39) defines an 'emotion recognition system' as one that identifies or infers emotions or intentions on the basis of biometric data. The recruitment reading comes from Commission Guidelines on prohibited AI practices, C(2025) 5052 final of 29.7.2025 (non-binding): 'Using emotion recognition AI systems during the recruitment process is prohibited.' Guidelines source: https://ec.europa.eu/newsroom/dae/redirection/document/112367.
- EUIn force
- What it requires
- Moves the start date of the AI Act's high-risk rules (Chapter III, Sections 1 to 3) for Annex III systems, including recruitment and selection AI, from 2 August 2026 to 2 December 2027. It also rewrites the AI literacy duty as taking measures to support staff AI literacy, and adds a safeguarded basis for processing special-category data for bias detection.
- Applies to
- Providers and deployers of AI systems under Regulation (EU) 2024/1689, including those of Annex III employment and recruitment systems.
2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III
Primary source: publications.europa.eu Confirmed
Notes
Published in the OJ on 24.7.2026. Article 4 says it enters into force on the third day after publication, which is 27 July 2026; the text itself also uses 27 July 2026 as the date from which Articles 102 to 110 apply. Annex I product-related high-risk systems now apply from 2 August 2028. Article 111(2) is amended so that legacy high-risk systems are covered only if their design changes significantly after the Chapter III application date. New Article 4a(2) lets deployers of high-risk systems process special-category data for bias detection under the listed safeguards, and states that it does not create any obligation to carry out bias detection. Articles 26 and 86 and Annex III point 4 are not amended. EUR-Lex blocked automated fetching, so the text was loaded from the Publications Office. ELI: http://data.europa.eu/eli/reg/2026/1744/oj.
- EUIn force
- What it requires
- People have the right not to be subject to a decision based solely on automated processing, including profiling, with legal or similarly significant effects, unless necessary for a contract, authorised by law with safeguards, or based on explicit consent. In contract and consent cases, controllers must offer at least human intervention, a chance to state a view, and to contest.
- Applies to
- Controllers subject to the GDPR that take solely automated decisions, including profiling, about individuals such as job applicants.
right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her
Primary source: publications.europa.eu Confirmed
Notes
Recital 71 names 'e-recruiting practices without any human intervention' as an example. Articles 13(2)(f), 14(2)(g) and 15(1)(h) require telling people that such automated decision-making exists, with meaningful information about the logic involved and its significance and envisaged consequences. Article 22(4) restricts decisions based on special-category data. Application date 25 May 2018 (Article 99(2)). A pending Commission proposal, COM(2025) 837, would rewrite Article 22(1)-(2) (see separate row). EUR-Lex blocked automated fetching, so the text was loaded from the Publications Office.
- EUIn force
- What it requires
- Non-binding guidance: human involvement must be meaningful, not a token gesture, by someone with authority and competence to change the decision. Decisions denying someone an employment opportunity could be 'similarly significant'. Its example says fully automated sifting of exceptionally high application volumes to shortlist may be necessary for entering a contract, if no less intrusive means exist.
- Applies to
- Controllers subject to the GDPR that use automated decision-making or profiling, including in recruitment.
decisions that deny someone an employment opportunity or put them at a serious disadvantage
Primary source: ec.europa.eu Confirmed
Notes
This is non-binding guidance, so effective_date is left null. The EDPB endorsed it on 25 May 2018 in Endorsement 1/2018 (https://www.edpb.europa.eu/sites/default/files/files/news/endorsement_of_wp29_documents_en_0.pdf). The EDPB's guideline page still listed it as the current guideline on 2026-09-28, and no replacement was found. It predates the AI Act. The applicant-shortlisting example appears in its discussion of the Article 22(2)(a) contract exception.
- EUEnacted, not yet in force
- What it requires
- Digital labour platforms must not use automated monitoring or decision-making systems to process certain personal data (such as emotional or psychological state, private conversations, or data to infer ethnic origin, health or sexual orientation), from the start of recruitment or selection. They must tell people undergoing recruitment, before it starts, about automated systems used in that procedure.
- Applies to
- Digital labour platforms as defined in the Directive, through national transposing law; protects persons performing platform work, including during recruitment or selection.
This Article shall apply to all persons performing platform work from the start of the recruitment or selection procedure.
Primary source: publications.europa.eu Confirmed
Notes
Status reflects when platforms must comply, not the Directive's own legal status. The Directive entered into force on the twentieth day after OJ publication (Article 31), but its obligations reach platforms through national law, which Member States must bring into force by 2 December 2026 (Article 29(1)). Article 7(3) extends the data limits to any automated systems taking or supporting decisions that affect persons performing platform work. National transposition status was not checked. It applies only to digital labour platforms, not to employers in general.
- EUProposed
- What it requires
- Would replace GDPR Article 22(1)-(2) so that a solely automated decision with legal or similarly significant effects may be taken only where necessary for a contract, regardless of whether it could be taken otherwise than by solely automated means; authorised by law with safeguards; or based on explicit consent.
- Applies to
- Would apply to controllers subject to the GDPR that take solely automated decisions, including in hiring.
a contract between the data subject and a data controller regardless of whether the decision could be taken otherwise than by solely automated means
Primary source: publications.europa.eu Confirmed
Notes
This is a proposal, not law. The GDPR amendments are in Article 3 of the proposal (Article 1 amends Regulation (EU) 2023/2854, the Data Act), and point 7 of Article 3 replaces GDPR Article 22(1)-(2). The European Parliament Legislative Observatory (https://oeil.secure.europarl.europa.eu/oeil/en/procedure-file?reference=2025/0360(COD)), checked 2026-09-28, shows the stage as 'Awaiting committee decision', with a committee draft report dated 22/06/2026 and amendments tabled 27/07/2026. Secondary sources say a Council compromise text dropped the Article 22 change; no Council primary source was loaded to confirm this. This proposal is separate from the adopted Regulation (EU) 2026/1744 (Digital Omnibus on AI).
- US: New York CityIn force
- What it requires
- Employers and employment agencies may not use an automated employment decision tool to screen candidates or employees unless it was bias-audited within the prior year and a results summary is posted publicly. NYC-resident candidates must get notice at least 10 business days before use, naming the qualifications assessed, and may request data-source and retention information.
- Applies to
- Employers and employment agencies using an AEDT to screen candidates for employment or employees for promotion in New York City.
it shall be unlawful for an employer or an employment agency to use an automated employment decision tool to screen a candidate or employee for an employment decision unless
Primary source: legistar.council.nyc.gov Confirmed
Notes
The law text says it 'takes effect on January 1, 2023.' DCWP says it began enforcing the law and its rule on July 5, 2023 (nyc.gov/site/dca/about/automated-employment-decision-tools.page). The notice must let a candidate request an alternative selection process or accommodation. Data-source and retention information must be given within 30 days of a written request. Penalties: up to $500 for a first violation and $500 to $1,500 for each later one, and each day of use is a separate violation. Local Law 25 of 2026 (Int. 1066-A of 2024) only orders a city study of how these tools affect municipal employees; it does not amend §§ 20-870 to 20-874. This row was checked against the enacted law, not the current consolidated Administrative Code, so later amendments are not ruled out.
- US: New York CityIn force
- What it requires
- Bias audits must calculate selection or scoring rates and impact ratios by sex, race/ethnicity and intersectional categories, using historical data unless insufficient. Before use, the audit date, results summary and distribution date go on the employment website, kept 6 months after last use. Candidate notice goes on the website, in the job posting, or by mail/email, 10 business days ahead.
- Applies to
- Employers and employment agencies using an AEDT in New York City, and the independent auditors conducting their bias audits.
a bias audit must, at a minimum: (1) Calculate the selection rate for each category; (2) Calculate the impact ratio for each category;
Primary source: nyc.gov Confirmed
Notes
The rule text was read in the official City Record of April 6, 2023 (pp. 1495-1498). The effective_date shown is the enforcement start that DCWP states for 'this law and rule'; the City Record notice does not give the rule's formal effective date. Test data may be used only when there is not enough historical data, and the summary must then explain why. Notice must say how to request an alternative selection process or accommodation, but the rule says nothing in it requires offering an alternative process. For promotion candidates, notice can go in a written policy. DCWP also publishes an FAQ (nyc.gov/assets/dca/downloads/pdf/about/DCWP-AEDT-FAQ.pdf).
- US: ColoradoEnacted, not yet in force
- What it requires
- Before a covered ADMT materially influences an employment decision, deployers give clear notice (a prominent public notice at points of interaction suffices). Within 30 days of an adverse outcome: plain-language description of the ADMT's role, how to get details, and rights. On request, data correction and meaningful human review where commercially reasonable. Developers document intended uses and limits; three-year records.
- Applies to
- Deployers and developers doing business in Colorado; 'consumer' includes any employee, and any job applicant who is a Colorado resident.
PRIOR TO A DEPLOYER USING A COVERED ADMT TO MATERIALLY INFLUENCE A CONSEQUENTIAL DECISION, THE DEPLOYER SHALL PROVIDE A CLEAR AND CONSPICUOUS NOTICE TO A CONSUMER
Primary source: leg.colorado.gov Confirmed
Notes
Signed May 14, 2026. The act 'takes effect January 1, 2027' and applies to consequential decisions made on or after that date. Its rulemaking provisions took effect on passage: the Attorney General must adopt rules on adverse-outcome disclosures and consumer rights by January 1, 2027. Only the Attorney General enforces it, under the Colorado Consumer Protection Act, and it creates no new private right of action. 'Materially influence' excludes incidental, trivial or clerical uses. The correction right does not cover opinions, predictions, scores or protected evaluations. The HIPAA covered-entity exemption does not apply to employment decisions. The session-law chapter number comes from the leg.colorado.gov bill page.
- US: ColoradoProposed
- What it requires
- Draft rules would require written adverse-outcome disclosures giving specific principal reasons, identifying inferences and underlying personal data, disclosing the consumer's score when a score was a principal reason, and stating automatic-denial factors or missing information. On request, each personal-data source must be named. An FCRA adverse action notice can suffice if it contains the required content.
- Applies to
- Deployers and developers of covered ADMT under Colorado's ADMT law (SB26-189), including employers using it in hiring.
the disclosure shall state that the Adverse Outcome was based on a profile or risk or other score, shall disclose the Consumer’s score
Primary source: coag.gov Confirmed
Notes
This is a proposed draft and has not been adopted. It was filed August 11, 2026. coag.gov/ai says written comments are accepted from August 11 through October 26, 2026, and that comments sent by September 4, 2026 would inform a revised draft. The hearing notice sets the public hearing for October 26, 2026 at 10:00 AM. The draft says the rules 'shall become effective January 1, 2027'. It proposes disclosure by at least two methods where the deployer has two, and a 12-point minimum font for printed notices. Under draft Rule 6.6, sources of personal data must be identified by name when a consumer asks for more information; this is not part of the adverse-outcome disclosure itself. The draft also covers chatbot duties under HB26-1263, which do not concern hiring.
Method
How it is kept up to date
- Entries are written by hand in the lab's admin. An entry cannot be saved without a link to its source, the day it was checked and who checked it. A new entry starts as a draft that no visitor can see.
- An entry is published only once its reading is marked confirmed against the source. The one exception is a super admin publishing with a written reason, which is recorded; that entry is shown here as Unconfirmed.
- The date at the top of this page is the most recent of the entries' checked dates, not a review of the whole list. Each entry keeps its own date.
- There is no automatic feed from legislatures or regulators and no fixed review schedule. Entries are added and revised as the lab checks them, so always read an entry's own date.