GRC Careers: Governance, Risk and Compliance JobsGovernance · Risk · Compliance Careers

Home › AI Governance Insights › How the EU AI Act Changes Recruitment and Hiring

How the EU AI Act Changes Recruitment and Hiring

By F. Jay Hall, GRC Careers Reference Library · November 17, 2026 · Updated November 19, 2026 · 22 min read

Published November 17, 2026. Every legal claim was checked against the primary sources listed at the end, most recently on November 19, 2026. This is a reference resource, not legal advice.

Consider a compliance manager applying for a role in Paris. She has led bank examinations and remediated audit findings, but her last job title was “Program Director.” The employer’s screening tool ranks 200 applications. Her file lands at number 31. The recruiter opens the first 20, conducts six interviews and signs off on the hire. Who decided she would never get an interview?

That question is at the heart of the EU Artificial Intelligence Act. Recruitment AI can shape an employment decision long before anyone presses “reject.” Annex III, point 4(a), covers AI systems intended to be used for the recruitment or selection of people, and it names placing targeted job advertisements, analyzing and filtering job applications, and evaluating candidates. Point 4(b) follows the worker into the job, covering specified systems for promotion, termination, task allocation and performance monitoring. The final hiring decision remains with the recruiter. The ranking already decided whose file reached the desk. [1] [3]

The date that matters for hiring systems

Regulation (EU) 2024/1689 entered into force on August 1, 2024. In July 2026 the EU enacted Regulation (EU) 2026/1744, the Digital Omnibus on AI. Amended Article 113(c)(i) sets December 2, 2027 as the application date for Chapter III, Sections 1 through 3, for Annex III high-risk systems, apart from Article 6(5). These sections contain the classification rules, system requirements and duties for providers and deployers discussed here. August 2, 2028 is the date for a separate class of systems tied to Annex I product legislation. [1] [2]

Other duties already apply. The workplace emotion-recognition prohibition discussed below has applied since February 2, 2025. The GDPR already governs covered personal data and certain automated decisions. In some markets, employers must already tell applicants about hiring AI before they submit an application. Ontario, for example, puts that disclosure in the job ad itself. [1] [6] [17] [18]

For existing systems, Article 111(2) adds a transition rule. For a typical hiring system already placed on the market or put into service before the relevant Chapter III date, significant design changes from that date bring it within the new regime. The Article 5 prohibitions still apply. The same paragraph sets an August 2, 2030 backstop for systems intended for public authorities. Record when the system entered service and what changed in each release. [1] [2]

The practical starting point is the vacancy. Follow it from advertisement to rejection, identify who controls each step, and ask what the candidate would have seen or experienced if the tool had been turned off.

Which recruitment tools qualify

From publishing a vacancy to choosing its audience

Imagine that an employer posts the Paris vacancy on a public job board. Every reader can open the same listing. A visitor chooses “Paris” and sorts openings by posting date. Those facts alone establish neither an AI system under Article 3(1) nor a high-risk hiring use. Now suppose a vendor predicts who is likely to apply and selectively delivers that vacancy to people who resemble the employer’s previous hires. Qualified people may never learn about the opening. AI intended for that type of recruitment advertising is expressly listed in Annex III. Saving applications on a different website does not change what the advertising tool does. [1] [3]

At the other end of the process, a candidate may privately use a writing assistant to sharpen a cover letter. Ask where the output goes. A draft the candidate controls serves a different purpose from a suitability score sent to an employer. If a platform starts selling rankings to recruiters, classify that feature separately, even if both features use the same underlying model. Article 2 can reach a provider established outside the EU that places a system on the EU market, and a third-country provider or deployer where the system’s output is used in the EU. A vendor’s billing address cannot settle the question. [1]

The narrow exception and the profiling rule

Article 6(3) allows an Annex III use to fall outside high-risk classification where it poses no significant risk to health, safety or fundamental rights, including because it does not materially influence the decision. It identifies four possible functions. The system may perform a narrow procedural task, improve work a human has already completed, detect patterns in prior decisions under the provision’s human-review safeguards, or prepare for a relevant assessment. An AI tool that organizes files for a reviewer may qualify. If it suppresses files it considers irrelevant, test the effect of that change. [4]

There is an overriding limit. An Annex III system that profiles individuals is always high-risk. A tool that evaluates personal characteristics to predict job performance cannot rely on the narrow-task exception merely because a recruiter can change its score. A provider relying on Article 6(3) must document its assessment before market placement or first use, register under Article 49(2), and provide the assessment to the national competent authority on request. [1] [4]

An interview feature that crosses a different line

An interview tool that analyzes what an applicant said and a tool that infers anxiety from facial movements or vocal cues raise different questions. Article 5(1)(f) prohibits AI emotion inference in the workplace, subject to the medical or safety exception. Article 3(39) supplies the biometric basis of emotion recognition. The Commission interprets workplace coverage to include candidates during selection. For an ordinary interview, an applicant’s consent cannot authorize this prohibited use. [1] [5]

Who is responsible for the shortlist

The vendor that supplies the system

A company that develops a hiring system, or has one developed, and markets it or puts it into service under its own name is generally its provider. For a covered Annex III hiring system, Articles 9 through 15 address risk management, data governance, technical documentation, logging, information for the deployer, human oversight, accuracy, robustness and cybersecurity. Articles 16 and 17 assign provider duties, including a quality management system. Under Article 43(2), the ordinary conformity assessment for Annex III point 4 uses Annex VI internal control. A customer asking for an independent EU bias-audit certificate would be asking for something different. [1] [7]

For the Paris vacancy, ask the provider which languages and job families it tested, where it found errors, and which version produced candidate 31’s score. If it tested English-language software-engineering vacancies, that evidence says little about French-language bank compliance work. Ask for the instructions and limitations that would lead a reasonable employer to switch off the ranking. This is a practical test of the documentation the customer receives. [1] [7]

The employer or agency that uses it

Under Article 26, the deployer must use the system in line with the provider’s instructions, appoint people with the competence, training, authority and support to oversee it, and monitor its operation. Where the deployer controls input data, it must ensure relevance and sufficient representativeness for the intended use. It must keep automatically generated logs under its control for an appropriate period of at least six months, unless applicable EU or national law provides otherwise. That rule concerns system logs. It supplies no blanket reason to keep every applicant’s résumé for six months. [8]

An employer deploying a high-risk system in the workplace must inform affected workers and their representatives before use. Article 26(11) separately requires notice to people where a covered Annex III system makes or assists decisions about them. Worker notice and applicant notice reach different people. A recruiter who can override a score only after opening a file needs a way to discover errors in files that never reach the interview queue. [7] [8]

When a customer becomes a provider

Under Article 25, an employer, recruiter or platform can assume provider duties by putting its brand on an existing high-risk system, subject to the article’s contractual qualification, by substantially modifying a high-risk system, or by changing another AI system’s intended purpose so it becomes high-risk. An agency adapting a general writing tool into a candidate scorer should assess its role before offering that service. Record the intended use, the change made, whose name appears on the product and who controls later updates. [9]

What the applicant can challenge

The Paris applicant learns that her file was screened. Three questions follow. Did anyone actually consider her experience? What role did the software play? Who has authority to put her back into consideration? Different provisions answer different parts of that conversation.

A decision made entirely by automation

GDPR Article 22 generally protects a person against a solely automated decision with legal or similarly significant effects. Recital 71 gives e-recruiting without human intervention as an example. Article 22(2) identifies exceptions for contract necessity, legal authorization with safeguards, or explicit consent. For the contract and explicit-consent routes, paragraph 3 requires safeguards that include human intervention, the opportunity to give one’s view and the ability to contest the decision. Convenience alone does not establish contract necessity. [6]

If nobody can see or rescue a below-cutoff application, investigate whether the rejection was effectively automatic. A reviewer who reads the underlying application, exercises independent judgment and has a genuine opportunity to change the outcome presents different facts. The AI Act can still classify the ranking tool as high-risk when it assists that reviewer. The GDPR test for a particular decision and the AI Act test for a system can yield different answers. [1] [3] [6]

An explanation when AI influenced the decision

Under AI Act Article 86, an affected person can seek a clear, meaningful explanation from the deployer of the AI system’s role and the main elements of the decision. Conditions matter. The deployer must have based the decision on the output of a covered Annex III system, and it must have legal or similarly significant effects that the person considers adverse to health, safety or fundamental rights. The article includes limits where other EU law already supplies the right and where lawful exceptions apply. Article 85 provides a route to complain to the relevant market surveillance authority. [1] [10]

“Your application was unsuccessful” gives candidate 31 no way to spot a factual error. A helpful response, if the record supports it, would explain that a ranking determined which files were reviewed, that the system failed to credit her bank-examination work, and who can reconsider the file. Avoid reconstructing a plausible-sounding reason after the fact. [8] [10]

Two different impact assessments

GDPR Article 35 requires a data protection impact assessment when processing is likely to pose high risk, including its specified case of systematic, extensive automated evaluation supporting significant decisions. AI Act Article 27 separately requires a fundamental rights impact assessment for specified deployers, including bodies governed by public law and private entities providing public services, with additional specified financial uses. An ordinary private employer does not acquire that second duty simply by purchasing hiring software. A public employment body may face both. Article 27(4) allows the assessments to complement one another. [6] [11]

What changes in the United States

Maryland and Washington regulate different uses of the same camera

In Maryland, Labor and Employment § 3-717 addresses the creation of a facial template during an applicant’s employment interview. An employer using facial recognition for that purpose needs the applicant’s consent through a signed waiver. The waiver must state in plain language the applicant’s name, interview date, consent to use facial recognition and whether the applicant read the waiver. Ask the vendor whether its software extracts a machine-readable facial pattern. A video recording by itself does not establish that it does. [12]

Washington’s RCW 43.386.030 asks a different question. When a state or local government agency uses facial recognition for a decision with legal or similarly significant effects, it must provide meaningful human review. The law expressly includes employment opportunities. Under RCW 43.386.010(7), the reviewer must be trained and have authority to alter the decision. A county hiring office should be able to name that reviewer. A private employer in Seattle cannot infer its own duties from a statute addressed to government agencies. [13]

Illinois and California reach beyond facial templates

For Illinois-based jobs, an employer that asks for recorded video interviews and analyzes them with AI must give advance notice, explain how the system works and the general characteristics it evaluates, and obtain consent. The Video Interview Act also limits sharing and requires deletion within 30 days of the applicant’s request, including instructions to recipients to delete copies. Separately, the Illinois Human Rights Act addresses discriminatory effects of AI in recruitment and other employment decisions and the use of ZIP codes as proxies for protected classes. Analyze the video and the selection decision separately. [14] [15]

California’s employment regulations address the selection procedure itself. Under 2 CCR § 11009(f), use of an automated decision system or selection criterion that discriminates on a protected basis is unlawful, subject to available defenses. Delivering a notice tells a candidate that a tool is in use. It leaves open whether the tool screens out qualified people on a protected basis. Review actual outcomes by job and selection stage. [16]

New York City and Colorado attach duties to the decision

In New York City, a covered employer or employment agency using an automated employment decision tool must obtain an independent bias audit within the preceding year, make a summary public, and give the required advance notices. The local notice period for covered city residents is ten business days. The city’s rule explains when an output substantially assists or replaces discretion, including when it outweighs every other criterion. Check how recruiters weight the score and which population the audit covers. [19] [20]

Colorado’s enacted SB 26-189 sets January 1, 2027 for its principal requirements. It covers specified automated technology that materially influences consequential decisions, including employment. Among other duties, deployers owe notice at interaction and a plain-language disclosure within 30 days following a covered adverse outcome. Affected people can request correction of inaccurate factual data and meaningful human reconsideration to the extent commercially reasonable. A team buying a ranker in 2026 has time to build those steps into the rejection process. [21]

Hiring disclosures and the right to challenge a decision

Ontario puts disclosure in the vacancy

Since January 1, 2026, Ontario’s Employment Standards Act § 8.4 requires a covered employer advertising a public vacancy to disclose in the posting if it uses AI to screen, assess or select applicants. O. Reg. 476/24 exempts employers with fewer than 25 employees on the posting date and defines which postings count. If a vendor screens for an employer, the candidate should still be able to find the disclosure in the ad. Separately, § 8.7 gives covered job posting platforms duties concerning fraudulent postings, including a reporting mechanism and a published policy. A platform can have obligations even where the employer controls candidate selection. [17] [18]

Brazil gives candidates a way to question a professional profile

Article 20 of Brazil’s Lei Geral de Proteção de Dados (LGPD) lets a person request review of a decision made solely through automated personal-data processing that affects their interests. The law expressly includes decisions defining a professional profile. On request, the controller must give clear, adequate information about the criteria and procedures used, subject to commercial and industrial secrecy. If information is withheld on that basis, the data authority may audit for discrimination. The current text guarantees a request for review without specifying that a human must perform every review. For a São Paulo applicant, make the route for questioning an automated rejection as easy to find in Portuguese as the vacancy. [22]

South Africa focuses on the consequences of profiling

Section 71 of the Protection of Personal Information Act restricts solely automated profiling decisions with legal or substantial effects on a person. Work performance is an explicit example of what a profile may evaluate. Its exceptions include certain contract-related decisions and decisions governed by law or a code with appropriate safeguards. Where the responsible party relies on the contractual safeguard in § 71(2)(a)(ii), it must give the person a chance to make representations and enough information about the logic to do so. If an employer automatically excludes a candidate based on a predicted performance score, identify the exception and the procedure that allows the candidate to be heard. [23]

Hiring systems in South Korea and Australia

South Korea addresses the system as well as the individual decision

The Korean AI Framework Act identifies consequential judgments such as hiring as a high-impact field, subject to its risk and significance test. Article 34 requires covered AI business operators providing high-impact AI or products or services using it to manage risks, put human oversight and user protections in place, and retain evidence. Separately, Personal Information Protection Act Article 37-2 addresses fully automated decisions, with a qualified objection right for significant effects on rights or duties and a route to request an explanation. For recruitment software sold in Seoul, the vendor’s evidence and the applicant’s route to an explanation answer different questions. [24] [25]

Australia adds a privacy policy obligation in December

On December 10, 2026, new Australian Privacy Principles 1.7 and 1.8 start applying to covered entities. Where a computer program uses personal information to make a decision, or performs a step substantially and directly related to it, and a significant effect on rights or interests is reasonably expected, the privacy policy must describe the kinds of information and decisions involved. This can reach software that assists a human. A covered recruiter should examine which fields feed the shortlist and describe the kinds of decisions the program helps make. [26] [27]

One practical difference appears when a recruiter reads a score and makes the final call. Keeping that employee in the process may matter to South Korea’s fully automated decision test. In Australia, substantial and direct computer assistance can still trigger a policy disclosure. In Ontario, using AI to assess candidates can be enough to require a statement in the ad. The same human approval step has three different legal consequences. [17] [25] [26]

Where the safeguard appears in the hiring process

Where the safeguard appears in the hiring process
Hiring step What the applicant or reviewer should be able to find
Opening the vacancyOntario. The covered employer’s AI screening disclosure in the public job ad. [17] [18]
Before the interviewMaryland. A signed, plain-language waiver if the employer will create a facial template. [12]
At the decisionWashington. A trained reviewer with authority to change a covered government facial-recognition decision. [13]
After an automated rejectionBrazil. A route to ask for review and for information about the criteria used. [22]

How to test whether a human is really deciding

Return to the Paris vacancy. Before a live shortlist goes to recruiters, give the hiring team two fictional applications with the same bank-examination and remediation experience. Give one the conventional title “Compliance Manager” and the other “Program Director.” Run them through the configured workflow, including the recruiter’s default screen. If one file disappears below the cutoff, ask the recruiter to find it without being told the candidate’s name.

Next, swap the order of equally qualified applications and repeat the exercise. Inspect what the recruiter sees, whether the full application is accessible, and whether moving a file back into consideration changes the actual interview queue. Record the tool version, cutoff, job criteria, observed difference and who approved any correction. Use fictional or appropriately controlled records for this exercise. A successful pairwise test is one piece of evidence. It cannot establish the selection rate or error rate for an entire applicant population.

If the problem occurs in production, identify which vacancies, versions and applicants could have been affected. Preserve the evidence needed to understand the error, assign a person with authority to suspend the ranking, and decide who can reconsider missed applications while interviews are still available. Under Article 26(5), where the deployer has reason to consider that use in line with instructions may present a risk within Article 79(1), it must inform the provider or distributor and the relevant market surveillance authority without undue delay and suspend use. A routine product bug and a risk meeting this test need different responses. [8]

A federal hiring case shows why that test matters

In its settled iTutorGroup case, the U.S. Equal Employment Opportunity Commission alleged that application software automatically rejected women aged 55 or older and men aged 60 or older. The agency said more than 200 qualified applicants were rejected. The companies agreed to pay $365,000. The settlement concerned age discrimination. A basic software rule can close the door before a human sees a qualified applicant. Ask for a demonstration of every automatic exclusion, including those outside the vendor’s advertised AI features. [28]

What this means at GRC Careers

GRC Careers publishes vacancies and educational resources. Our Trust and Governance Disclosure explains that we do not use algorithms to score, screen or rank applicants for employers. Applicants should use the application instructions in the listing. Please do not send us unsolicited résumés. If you want an individual résumé review, book and pay for a career appointment first. A paid appointment is for your own career advice. You can apply for a listed vacancy without booking one. An employer receiving your application through its own hiring system is responsible for explaining how it uses that system. [29]

Who can see the file that the system put in thirty-first place? For the applicant, what matters is whether someone can act on the answer while there is still an interview to offer.

We monitor the EU AI Act because a job board’s role changes if it begins targeting vacancies or screening people for employers. We publish career resources in Portuguese, including our guide to compliance jobs in Brazil [32]. Our policy review covers applicable requirements country by country across South America, Africa, Asia and Oceania.

For further reading, see The EU AI Act, Explained [30] and the EU AI Act Readiness Companion [31].

Each numbered marker in the text links to the entry below.

  1. EU Artificial Intelligence Act, consolidated text Regulation (EU) 2024/1689 as amended. Articles 2, 3, 5, 6, 9 to 17, 25 to 27, 43, 49, 85 to 86, 111 and 113. Annex III, point 4.
  2. Digital Omnibus on AI, Official Journal Regulation (EU) 2026/1744. Article 1 amends the application timetable. Published July 24, 2026.
  3. European Commission, employment use cases Illustrative guidance on job advertising, sourcing, matching and ranking. The statutory test appears in Annex III and Article 6.
  4. AI Act, Article 6 Classification, narrow exceptions, profiling and documentation. Read with Article 49(2).
  5. European Commission, guidelines on prohibited AI practices Interpretive guidance on Article 5(1)(f), especially paragraphs 253 to 255 on workplace scope and hiring.
  6. General Data Protection Regulation Regulation (EU) 2016/679. Articles 22 and 35 and recital 71 on automated decisions and data protection impact assessments.
  7. AI Act, Article 14 Design and operation of human oversight, including the ability to interpret, disregard or override outputs. See also Articles 9 to 17 in [1].
  8. AI Act, Article 26 Deployer duties. In particular, paragraphs 1 to 7 and 11 on use, oversight, monitoring, logs and notice.
  9. AI Act, Article 25 Changes in responsibility along the supply chain, including rebranding, substantial modification and a change in intended purpose.
  10. AI Act, Article 86 Conditions and limits on an individual’s right to an explanation. Article 85 in [1] covers complaints.
  11. AI Act, Article 27 Scope, timing and elements of the fundamental rights impact assessment. See paragraph 4 on overlap with a GDPR assessment.
  12. Maryland Code, Labor and Employment § 3-717 Facial templates during applicant interviews. Subsections (b) and (c) govern consent and waiver contents.
  13. Revised Code of Washington, chapter 43.386 Sections .010(7) and .030 define meaningful human review and identify covered government decisions.
  14. Illinois Artificial Intelligence Video Interview Act 820 ILCS 42/5, /10 and /15. Advance explanation and consent, limits on sharing and requested deletion.
  15. Illinois Human Rights Act 775 ILCS 5/2-102(L). Employment use of AI, discriminatory effects and ZIP-code proxies.
  16. California Civil Rights Council, approved employment regulations California Code of Regulations, title 2, §§ 11008.1 and 11009(f). Approval and regulatory text.
  17. Ontario Employment Standards Act, 2000 Part III.1, including § 8.4 on AI disclosures and § 8.7 on job posting platforms.
  18. Ontario Regulation 476/24 Rules and Exemptions Re Job Postings, §§ 1 to 2. Posting definitions and the employee threshold.
  19. New York City Local Law 144 of 2021 City Council legislation record and text. Administrative Code §§ 20-870 through 20-874.
  20. New York City adopted rule on automated employment decision tools 6 RCNY §§ 5-300 through 5-304. Definitions, audits, publication and notices.
  21. Colorado SB 26-189, signed act Colorado Revised Statutes §§ 6-1-1701 through 1705 and act § 5. Definitions, disclosures, review and application date.
  22. Brazil, Lei nº 13.709/2018, current text LGPD, Article 20, main paragraph and §§ 1 to 2. Câmara dos Deputados, Portuguese original.
  23. South Africa, Protection of Personal Information Act 4 of 2013 Section 71(1) to (3) on automated profiling, exceptions and representations. Statutory text.
  24. Republic of Korea, AI Framework Act Articles 2(4)(g), 33 and 34. High-impact AI and duties for covered business operators. English translation.
  25. Republic of Korea, Personal Information Protection Act Article 37-2 on automated decisions, objections, explanations and responsive measures. English translation.
  26. Australia, Privacy and Other Legislation Amendment Act 2024 Schedule 1, Part 15, items 88 to 89. New Australian Privacy Principles 1.7 to 1.9 and application provision.
  27. Australian Information Commissioner, automated decision transparency Regulator’s explanation of the December 10, 2026 commencement and privacy policy contents.
  28. U.S. Equal Employment Opportunity Commission, iTutorGroup Enforcement agency’s account of the age discrimination allegations and settlement, September 11, 2023.
  29. GRC Careers Trust and Governance Disclosure Publisher’s description of its own practices.
  30. The EU AI Act, Explained GRC Careers overview of the regulation and its implementation timeline.
  31. EU AI Act Readiness Companion GRC Careers reference and practice materials.
  32. Vagas de compliance no Brasil GRC Careers employment resource in Brazilian Portuguese.

Frequently Asked Questions

When do the EU AI Act's high-risk rules start applying to hiring systems?

December 2, 2027. Article 113(c)(i), as amended by Regulation (EU) 2026/1744, sets that date for Chapter III, Sections 1 through 3, for systems classified as high-risk under Article 6(2) and Annex III, which is where recruitment and hiring systems sit. August 2, 2028 is the date for the separate Annex I product class. Other duties already apply: the Article 5(1)(f) workplace emotion-inference prohibition has applied since February 2, 2025, and the GDPR already governs covered personal data and certain automated decisions.

Does a job board become a high-risk AI provider just by listing vacancies?

No. A public listing every reader can open, with filters for location and posting date, establishes neither an AI system under Article 3(1) nor a high-risk hiring use. What Annex III point 4(a) reaches is AI intended for the recruitment or selection of people, including targeted job advertising that decides who is shown a vacancy, and systems that analyze, filter or evaluate applications. A board's role changes if it starts targeting vacancies or screening people for employers.

Can a recruiter's final sign-off keep a ranking tool out of scope?

Not on its own. Article 6(3) can take an Annex III use out of high-risk classification where it poses no significant risk, but an Annex III system that profiles people is always high-risk, and a provider relying on Article 6(3) must document the assessment before market placement or first use and register under Article 49(2). Separately, GDPR Article 22 asks a different question about a particular decision, so the two tests can give different answers for the same workflow.

What can a rejected applicant actually ask for?

Three different things, from three different provisions. Under GDPR Article 22 and its paragraph 3 safeguards, a person facing a solely automated decision with significant effects can seek human intervention, give their view and contest the outcome. Under AI Act Article 86, a person affected by a decision a deployer based on the output of a covered Annex III system can seek a clear, meaningful explanation of the system's role and the main elements of the decision, subject to that article's conditions and limits. Article 85 provides a route to complain to the market surveillance authority.

Do the United States, Canada and other markets already require hiring AI disclosures?

Several do, and they attach the duty at different points. Ontario has required the disclosure in the public job ad since January 1, 2026. New York City requires an independent bias audit within the preceding year, a public summary and advance notice. Illinois requires advance explanation and consent for AI-analyzed video interviews. Maryland requires a signed waiver before a facial template is created in an interview. Colorado's SB 26-189 sets January 1, 2027 for its principal requirements. Australia's new Privacy Principles 1.7 and 1.8 start applying December 10, 2026.

Who's Hiring AI Governance Professionals?

Explore current openings in:

AI Governance Jobs · Responsible AI · AI Risk · AI Compliance · AI Audit · AI Policy · Privacy · Cybersecurity · Public Policy

Search the latest opportunities at GRC Careers › · nonprofit GRC jobs ›