How AI Resume Screening Works 2026 Explainer

AI resume screening is now a standard layer in most enterprise hiring workflows, yet the gap between what vendors promise and what the technology actually delivers remains wide. This guide breaks down exactly how AI resume screening works, from the moment a resume is uploaded to the moment a score is assigned, and maps the documented failure modes that cost recruiting teams qualified candidates. Whether you are evaluating a new screening tool, auditing an existing one, or trying to set realistic expectations for your hiring managers, understanding the mechanics matters before trusting the output.

What Is AI Resume Screening?

AI resume screening is the automated process of using artificial intelligence technologies, including machine learning algorithms and natural language processing, to evaluate, filter, rank, and shortlist job applicants based on predefined criteria. At its core, the system parses resume text into structured data, then scores each candidate against a job description using machine learning models. Most systems combine keyword extraction, semantic similarity matching, and ranking algorithms trained on historical hire data.

It is important to distinguish AI resume screening from the older, simpler applicant tracking system (ATS). A traditional ATS is primarily workflow software that organizes and stores job applications. AI screening is a layer built on top of it that attempts to evaluate candidate fit rather than just store records. In practice, many platforms market themselves as AI-powered when the underlying logic is still largely rules-based keyword filtering with a more modern interface. Understanding which category a given tool falls into is one of the most useful questions a talent acquisition buyer can ask during an evaluation.

Why AI Resume Screening Matters in 2026

Application volumes have grown sharply over the past several years. At 250 or more applications per role, manual resume review consumes dozens of recruiter-hours per hire and produces inconsistent results that let strong candidates slip through on keyword luck alone. A single job posting now attracts an average of 250 resumes, and at seven seconds per resume, working through that stack manually takes 30 hours. Most recruiters spend two to three hours on this bottleneck every day.

Adoption has followed volume. AI resume screening software went from 26% to 43% of HR teams between 2024 and 2025, and as of 2025, 97.8% of Fortune 500 companies use an applicant tracking system as the foundation of their hiring workflow. The technology has shifted from optional to operationally necessary for high-volume recruiting teams.

The compliance environment has also changed. The EU AI Act classifies recruitment screening tools as high-risk AI systems, with full enforcement beginning August 2026 and penalties up to €30,000,000 euros or 6% of global turnover for serious violations. In the United States, the EEOC has opened initiatives on algorithmic fairness in hiring, and New York City and Colorado are the only jurisdictions currently requiring mandatory auditing of AI hiring systems. These regulatory pressures are raising the stakes for teams that deploy screening tools without adequate oversight.

For talent acquisition leaders evaluating tools in this environment, speed is only part of the value proposition. Accuracy, explainability, and bias governance are equally important dimensions to score tools against.

How AI Resume Screening Actually Works: The Mechanics

Understanding the sequential process that happens between application submission and shortlist generation is essential for setting realistic expectations and identifying where errors accumulate.

Step 1: Resume Parsing

The process begins with resume parsing. The system ingests resumes in multiple formats, typically PDF, Word documents, and sometimes LinkedIn profiles, and extracts structured fields: job titles, company names, dates of employment, skills, education, and certifications. This step converts unstructured documents into searchable, comparable data that the scoring layer can work with.

Parsing accuracy is not uniform across all fields. Basic fields like name and email consistently hit accuracy rates of 0.99 or higher. Field-level accuracy for skill extraction in production resume parsers ranges from 0.75 to 0.85. That gap is where qualified candidates begin to disappear, because every downstream matching decision inherits the error rate of the weakest extracted field. Resumes with heavy graphics can experience parsing errors as high as 43%, compared to just 5% for plain-text formats.

Step 2: Knockout Filtering

Before any AI scoring takes place, most systems run a binary knockout filter based on pre-screening questions. Screening questions covering work authorization, minimum years of experience, and location remove anyone who fails the basics before a resume is read at all. This is still the largest single source of fast rejections in most pipelines. The fast rejections candidates experience are mostly caused by knockout questions and parsing failures, not by an AI model judging resume content.

Step 3: Semantic Matching and Scoring

Modern AI screening tools go beyond keyword matching. Using NLP models, they interpret the meaning behind resume language. A candidate who writes “architected microservices infrastructure” can be matched to a role requiring “backend system design” because the model recognizes the semantic relationship between those phrases. A keyword filter, by contrast, might reject a highly qualified candidate because they wrote “machine learning engineer” instead of “ML engineer.”

The parsed, semantically understood resume is then scored against the specific criteria set for the role: required experience level, must-have skills, preferred industries, education requirements, and defined deal-breakers. Each candidate receives a score reflecting how closely they align with job requirements. ATS platforms calculate scores by weighing several factors, with keyword matching contributing roughly 40 to 60% of the score, formatting accounting for 20 to 30%, section organization adding another 15 to 25%, and skills alignment making up roughly 10%. Machine learning models then normalize these scores on a scale of 0 to 100, with resumes scoring above 80 typically advancing to recruiters and those scoring below 60 often automatically rejected.

Step 4: Ranking and Shortlisting

Candidates are ordered from highest to lowest match score, and those below a certain threshold or missing must-have criteria are automatically filtered out. The highest-ranking candidates are shortlisted for recruiter review. In a well-configured system, the AI layer typically summarizes and ranks the resumes that already survived parsing and the knockout filter. A human recruiter then reviews that shortlist rather than the full applicant pool.

The overall workflow follows a consistent end-to-end path: job description, resume intake, parsing, feature extraction, scoring and ranking, human review, interview, and hire. AI augments several stages within that workflow, including automated parsing, semantic skill matching, ranking by composite fit scores, and routing to reviewers when model confidence is low.

Common Challenges in AI Resume Screening and Where Systems Fail

The mechanics described above work well under ideal conditions. In practice, several failure modes are well-documented, and the consequences compound at scale. A flawed human decision may affect a single candidate. A flawed algorithm can quietly exclude thousands before anyone notices.

Parsing Failures on Non-Standard Formats

Visually formatted resumes that designers, creative directors, and UX professionals often use consistently underperform in AI parsing. The visual layout that makes them compelling to a human reader makes them difficult for text-extraction systems. Text embedded in graphics, custom fonts that do not parse cleanly, and multi-column layouts that disrupt reading order all reduce parsing accuracy. Older parsers using rigid pattern-matching fail on two-column layouts in seven of eight ATS platforms. The result is a specific irony: the candidates most likely to have visually impressive resumes are the candidates whose resumes are most likely to be misprocessed by AI screening systems.

Date format inconsistencies create separate parsing errors. ATS systems expect standardized date formats, and variations between “Sept 2023,” “September 2023,” and “09/2023” can cause fields to be misread or dropped entirely.

Synonym and Vocabulary Gaps

Keyword-based filters match token presence and break on synonyms and context. A candidate listing “account management” gets screened out of a posting requiring “client relationship management,” even when the work is identical. Semantic matching systems reduce this problem by evaluating meaning rather than exact strings, but they introduce their own failure modes at the scoring threshold boundary. At a 0.85 similarity cutoff, a product manager describing “care pathway optimization” may not match a posting asking for “process improvement,” even though the meaning overlaps significantly.

The synonym gap problem is well-documented and measurable. AI-driven skill extraction in modern ATS platforms carries a 15 to 25% error rate according to current parsing benchmarks. That error rate, compounded across rising application volumes, means even carefully configured intake filters are silently discarding qualified candidates at scale.

Career Changers and Non-Linear Paths

AI screening models score candidates based on matching their past experience to the target role profile. Career changers, by definition, have past experience in different roles. That experience contains transferable skills that do not map cleanly to the new role’s keywords, and most models are not equipped to bridge that gap reliably. AI models struggle with non-linear career paths, career re-entry after caregiving gaps, or skills developed outside formal employment. A Harvard Business School study found that 71% of surveyed employers said their screening tools filtered out resumes with employment gaps, even when the gap was for caregiving or education.

Training Data Bias

AI screening models score candidates based on patterns in their training data. When that data reflects historical hiring decisions made by humans, it encodes and amplifies the biases present in those decisions. AI systems trained on biased historical data amplify that bias at scale and at speed, affecting far more candidates than a single biased human recruiter would. Research from the University of Washington found that AI models ranked resumes with White-associated names more often than those with Black-associated names, even when qualifications were identical. Simulations by the Brookings Institution on large language model resume screening similarly show systemic bias based on inferred demographic signals.

AI screening does not eliminate bias; it changes the bias profile, reducing keyword and formatting bias while potentially introducing model bias when training data is skewed. This makes independent audits a compliance requirement, not an option.

Black-Box Scoring and Explainability Problems

Many AI systems operate as black boxes where decisions are made but the reasoning is unclear. When a recruiter cannot explain why an AI ranked one candidate above another, the decision becomes difficult to defend. This is not just an ethical concern. Under Article 22 of GDPR, candidates have the right not to be subject to solely automated decisions with significant effects. A screening system that auto-rejects without any human review trail is non-compliant in the EU. Under the EU AI Act, from August 2026, vendors and deployers must complete mandatory risk assessments, maintain technical documentation, ensure human oversight mechanisms, and disclose AI use to candidates.

Resume Gaming and AI-Generated Content

Resume coaching services teach candidates how to optimize for ATS keywords, a practice known as ATS optimization or ATS hacking. This inflates keyword match rates without reflecting actual skill, degrading the quality signal in the scoring layer. Recent research found that large language models prefer AI-written resumes over human-written ones by a margin of 67 to 82%, an artifact that has nothing to do with candidate quality. Hidden keywords and keyword stuffing are now actively detected by some platforms and can flag an application as manipulative. The practical effect is an arms race between candidates optimizing for AI and AI systems trying to detect that optimization.

What to Look for in an AI Resume Screening Tool

Knowing the failure modes clarifies what to ask when evaluating a tool. The criteria below reflect what separates screening systems that produce reliable shortlists from those that generate a false sense of efficiency while losing qualified candidates.

Parsing Fidelity and Format Handling

Test parsing accuracy before committing to any platform. Feed the system a representative sample of resumes in the formats your actual candidates submit, including PDFs with multi-column layouts, Word documents with unusual section headers, and text-heavy resumes without graphics. Confirm that structured fields, particularly skills, job titles, and employment dates, are extracted correctly. A system with 0.75 skill extraction accuracy is silently miscategorizing one in four skill mentions. That error compounds across every downstream scoring decision.

Semantic Matching Depth

Keyword matching and semantic matching are not equivalent. A tool that only checks for exact keyword presence will fail on synonym variation and natural language variation in resumes. Test whether the system correctly matches semantically equivalent phrases. A candidate who describes building distributed systems should score well against a role requiring cloud infrastructure experience, even without exact keyword overlap.

Scoring Explainability

Every shortlist generated by an AI screening system carries legal and operational risk if the scoring logic cannot be interrogated. Look for platforms where each candidate score comes with an explanation of which criteria were met and where gaps exist. That transparency allows recruiters to identify when the model is rewarding resume formatting rather than actual qualifications, and it provides an audit trail if a hiring decision is challenged. Under the EU AI Act, explainability is a regulatory requirement, not a product feature.

Bias Audit Documentation

Any reputable vendor should be able to provide a bias audit showing disparity impact analysis across gender and ethnicity at minimum. If a vendor cannot produce one, that is a material risk signal. Before deploying any scoring model to filter candidates, run it on a sample of 50 to 100 resumes and analyze the outputs by demographic proxy where visible from name, university, or previous employer. Test for adverse impact: check whether resumes with female names score lower, whether bootcamp graduates are penalized relative to university graduates, and whether employment gaps carry disproportionate scoring penalties.

Human Override Capability

A well-designed screening system surfaces scores and recommendations for human review rather than making final decisions autonomously. Verify that the platform includes a documented human override flow, that a recruiter can see the reasoning behind a score, and that overrides are logged with a recorded reason. If the override process is clunky or undocumented, it will not hold up to regulatory scrutiny and it signals that the vendor’s design philosophy treats the AI as a decision-maker rather than a decision-support tool.

Integration with the Existing ATS

A screening layer that does not connect cleanly to the ATS creates data entry redundancy and breaks the audit trail. Confirm that scores, ranking data, and candidate notes flow directly into the ATS without requiring manual export and re-upload. Integration depth also affects compliance: if screening data lives outside the ATS, it becomes difficult to produce a complete record of the decision-making process if a hiring outcome is challenged.

How Recruiting Teams Use AI Resume Screening in Practice

The best-performing implementations of AI resume screening share a common characteristic: they treat the AI as a filter and summarizer, not as a decision-maker. The following strategies reflect how talent acquisition teams get measurable value from screening technology while managing its documented failure modes.

Use ML parsing universally, scoring selectively. The return on eliminating manual data entry through automated parsing is clear and the bias risk is minimal. Every candidate who applies should have their resume parsed automatically into a structured record. Scoring, however, should be restricted to high-volume roles with 100 or more applications and clearly defined criteria where the target profile is stable and consistent. Avoid using automated scoring for senior roles, executive roles, roles where the criteria are evolving, or roles where past hiring data reflects demographic patterns you are actively trying to change.

Calibrate the model before deploying it. Feed the AI model 50 to 100 resumes that your team has already manually reviewed and scored. This calibration step ensures the model’s scoring aligns with your actual hiring decisions rather than a generic benchmark. After calibration, re-run a bias audit to verify that the adjusted scoring weights do not introduce new disparate impact.

Apply a tiered review structure rather than binary filters. A well-configured screening workflow advances the top 15 to 20% of scored candidates automatically to the phone screen, routes the middle 20 to 35% to human review, and auto-rejects only the bottom 50 to 60%. Switching from binary knockout filters to weighted scoring tiers with monthly rejection audits reduces false-negative candidate loss by 30 to 40%.

Audit rejection outputs regularly. Most recruiting teams never check what their ATS actually rejects. Reviewing 25 to 30 declined applications each month surfaces synonym gaps in job descriptions, identifies where the scoring model is penalizing non-traditional candidate profiles, and catches demographic disparities before they accumulate into a legal exposure. This audit step is a basic operational requirement, not an advanced practice.

Maintain candidate communication standards. AI screening does not eliminate the obligation to communicate with applicants. Candidates should be notified when AI is being used in the screening process. This is legally required in multiple jurisdictions and is a baseline best practice regardless of jurisdiction. Candidates who do not advance deserve a timely, professional response, and eliminating that contact in the name of efficiency damages the employer brand and removes the opportunity to build a pipeline relationship for future roles.

Keep the human review layer visible and documented. The AI layer in a compliant screening workflow summarizes and ranks. A human recruiter reviews the shortlist and owns the decision to advance a candidate. That review step needs to be documented with a timestamped record, not implied. Under both GDPR Article 22 and the EU AI Act, a screening system that auto-rejects without a documented human review trail is non-compliant.

Best Practices and Expert Guidance for AI Resume Screening

The following practices reflect the approaches that consistently reduce false negatives, manage bias risk, and produce defensible shortlists.

Write structured job descriptions before configuring the AI. AI screening quality is a direct function of job description quality. Poorly structured or vague job descriptions produce vague scoring criteria, which produce unreliable rankings. Before configuring any screening tool, work with the hiring manager to define must-have criteria, preferred criteria, and explicit deal-breakers. When criteria are explicit, the AI model’s scoring is transparent and defensible.

Map synonym coverage before deploying scoring criteria. Every required skill in the job description should map to at least two or three synonym variants in the screening configuration. A candidate describing “strategic account growth” should not score below a candidate who wrote the exact phrase from the job posting. Mapping synonyms requires a one-time effort per role family and measurably reduces false negatives on vocabulary variation.

Request model cards from vendors. A model card is documentation of a machine learning model’s intended use, performance benchmarks, training data characteristics, and known limitations. Requiring vendors to provide model cards during procurement is a straightforward quality signal. Vendors who cannot or will not provide them are not prepared for the regulatory environment that applies from August 2026 onward.

Restrict high-stakes scoring to roles where the criteria are stable. AI scoring models perform most reliably when the target profile is consistent across hiring cycles. Technical roles with clearly defined skill requirements are the strongest use case. Senior leadership roles, creative roles, and roles where cultural contribution matters as much as credential matching are poor fits for automated scoring as the primary filter.

Test for adverse impact before and after deployment. The 4/5ths rule from US employment law is a basic adverse-impact test: if a protected group’s selection rate is less than 80% of the group with the highest selection rate, adverse impact is indicated. Most AI resume screening vendors do not publish results against this test. Running it internally on a sample of scored resumes before go-live is a basic due diligence step that protects the organization regardless of what the vendor provides.

Treat scoring thresholds as variables, not fixed settings. The score threshold at which candidates advance or are rejected should be reviewed after each hire cycle. Rigid scoring thresholds may filter out qualified candidates when the candidate pool composition changes or when the hiring criteria are adjusted. A threshold that worked well for a previous cohort of applicants may produce materially different results for a current cohort.

Advantages and Benefits of AI Resume Screening for Talent Acquisition

With the failure modes documented above, the benefits of AI screening tools are real and measurable, provided the tools are selected and configured with the care the category requires.

Processing speed at scale. AI processes 1,000 or more resumes in under 30 minutes compared to 40 or more hours of manual review. A role that previously required a full day of manual review generates a shortlist in minutes. That speed advantage compounds across a high-volume hiring program and frees recruiter time for work that requires human judgment: structured interviews, stakeholder alignment, and offer negotiation.

Consistency. When properly configured, AI recruitment platforms apply identical criteria to every candidate. Human recruiters, despite best intentions, experience decision fatigue. Candidates reviewed at the end of a long day often receive less thoughtful consideration than morning applicants. AI does not get tired, and it applies the same scoring logic to the first resume and the thousandth.

Reduced formatting and vocabulary bias. AI screening can reliably identify resumes containing specific skills or qualifications even when they are laid out differently. This keeps the focus on candidate qualifications rather than formatting choices or font selection, which can disadvantage candidates who lack access to professional resume writing resources.

Structured data for pipeline analysis. Parsed resume data, when stored correctly in an ATS, enables downstream analytics that manual review cannot produce. Recruiters can analyze the composition of their applicant pool, identify stages where qualified candidates are dropping out, and measure whether job description changes are attracting different candidate profiles. That analytical layer is only as good as the parsing accuracy upstream.

Scalability without proportional headcount increases. A software company that needs to onboard 100 support technicians in 30 days cannot scale its recruiting team in that timeframe. AI screening handles the initial filter without adding agency costs or recruiter headcount, letting the organization respond to demand spikes while keeping the human review capacity focused on the most important decisions.

How to Evaluate AI Resume Screening Tools

When reviewing AI hiring software, scoring each tool on a consistent rubric helps create meaningful comparisons. For AI resume screening tools, that rubric should cover parsing fidelity across multiple file formats, semantic matching depth, scoring explainability, bias audit transparency, human override capability, and ATS integration quality.

A useful scoring framework reflects the mechanics and failure modes documented in this guide. A tool that processes resumes quickly but produces explainability-free scores, lacks a documented human override flow, or cannot produce a bias audit report should score materially lower than a tool with moderate processing speed and full transparency. Speed without accuracy and accountability is not a benefit in a category where a flawed algorithm can quietly exclude thousands of qualified candidates.

In independent testing, the tools that perform consistently well share three characteristics. First, their parsing layer handles non-standard formats without significant accuracy degradation. Second, their scoring logic is exposed to the recruiter, not hidden behind a summary label. Third, they support human override as a first-class workflow feature, not an afterthought. Tools that meet all three criteria are the ones that hold up when hiring decisions are scrutinized by a recruiting operations leader, a legal team, or a regulator.

Buyers evaluating tools in this category should use the failure modes in this guide as their test plan, not the vendor’s demo script. Feed the tool formatted resumes, visually designed resumes, and resumes from career changers. Check whether career gaps are penalized. Verify that scores come with explanations. Ask for the bias audit report before signing a contract.

The Future of AI Resume Screening

AI resume screening is not a solved problem. The technology is improving, but the gap between what is marketed and what is measured remains meaningful. Several trends are reshaping the category heading into the second half of 2026 and beyond.

Large language model-based screening is spreading quickly but is not yet universal. Most “AI screening” in production today is still keyword and rules logic with a more modern interface. True LLM-based screening adds semantic depth but introduces new failure modes, including model preference for AI-written resume content and sensitivity to prompt configuration that most vendors have not fully characterized. Agentic AI systems that work toward defined goals, such as screening, ranking, and summarizing candidates using contextual data, represent the next architectural shift. The practical implications for recruiting teams are still emerging.

Regulatory pressure will accelerate the demand for explainability and audit trails. The EU AI Act’s August 2026 enforcement date, combined with EEOC scrutiny in the US and emerging state-level legislation, means that the compliance posture of AI screening tools is becoming a direct input into vendor selection decisions. Teams that have not yet asked their vendors for bias audit documentation, model cards, and human oversight architecture should start now.

Skills-based hiring is the structural trend most likely to reshape what AI screening measures. As more employers move away from credential-based filtering toward demonstrated competency evaluation, the resume itself becomes a less reliable signal. That shift does not reduce the role of AI screening; it changes what the AI should be measuring. Tools that integrate screening with skills assessments and structured interview scoring will be better positioned than those that treat the resume as the primary evidence of candidate fit.

For talent acquisition teams building or updating their screening stack, the actionable steps are: audit what your current tools actually reject, request bias documentation from every vendor in your stack, confirm that human override is documented and compliant, and define your scoring criteria at the job description level before configuring any AI layer. The technology is a multiplier. What it multiplies depends on the quality of the inputs and the oversight applied to the outputs.

FAQs About AI Resume Screening

What is AI resume screening?

AI resume screening is the use of machine learning and natural language processing to automatically parse, score, and rank job applicants based on predefined role criteria. The system converts resume text into structured data, compares that data against the job description, assigns a match score, and ranks candidates so recruiters review a prioritized shortlist rather than an unsorted stack. Evaluating AI screening tools on parsing accuracy, scoring explainability, bias audit transparency, and human override capability helps talent acquisition buyers identify which tools actually deliver on those mechanics.

Why do recruiting teams need AI resume screening tools?

At 250 or more applications per role, manual screening consumes 30 or more hours of recruiter time and produces inconsistent results driven by fatigue and unconscious bias. AI screening processes that same volume in under 30 minutes and applies identical criteria to every resume. The business case is strongest for high-volume roles with stable, well-defined criteria. The catch is that speed does not equal accuracy. AI-driven skill extraction carries a 15 to 25% error rate in production systems, which means teams need to audit rejection outputs and configure synonym coverage to avoid losing qualified candidates in the process.

What is resume parsing and why does it matter?

Resume parsing is the first step in AI resume screening: the system extracts structured fields, including job titles, skills, employment dates, education, and certifications, from unstructured resume documents. Every downstream scoring decision depends on parsing accuracy. Basic fields like name and email parse reliably, but skill extraction accuracy in production parsers ranges from 0.75 to 0.85, meaning one in four to one in six skill mentions may be miscategorized or missed entirely. Visually formatted resumes can experience parsing error rates as high as 43%. Testing parsing fidelity across file formats and resume styles should be part of every AI screening tool evaluation.

How accurate is AI resume screening?

Reported accuracy rates for AI resume screening range from 70 to 85% depending on the tool, the role type, and the quality of the job description used to configure the scoring criteria. That range means a non-trivial share of scored resumes are ranked incorrectly relative to actual candidate fit. Accuracy degrades further for non-traditional career paths, career changers, and candidates whose resume vocabulary does not align with the exact language in the job description. Semantic matching systems outperform keyword-only systems, but both have documented false-negative rates. Accuracy claims from vendors should always be tested against a representative sample of real candidate resumes before deployment.

How does ATS resume scoring work?

ATS resume scoring works by weighing several signals extracted from the parsed resume against the job description. Keyword matching typically contributes 40 to 60% of the score, formatting and section organization account for 20 to 30%, and skills alignment makes up the remainder. Machine learning models normalize these weighted inputs into a score on a scale of 0 to 100. Resumes scoring above 80 typically advance to recruiter review; those scoring below 60 are often automatically rejected. The specific weights vary by platform and by how the tool has been configured for a given role. A recruiter who has not verified the weight configuration and tested for adverse impact is operating without full visibility into what the score actually measures.

Where does AI resume screening fail most often?

The most consistent failure modes are parsing errors on non-standard resume formats, synonym and vocabulary gaps between candidate language and job description language, systematic penalization of employment gaps and non-linear career paths, and training data bias that encodes historical hiring patterns into scoring outputs. Recent research also shows that large language models used for screening prefer AI-generated resume content over human-written content by 67 to 82%, introducing an artifact entirely unrelated to candidate quality. Documenting these failure modes as part of vendor evaluations gives buyers a clear picture of where each tool’s accuracy degrades.

What are the compliance requirements for AI resume screening in 2026?

The compliance landscape in 2026 is more demanding than in prior years. The EU AI Act classifies recruitment screening tools as high-risk AI systems under Annex III, with full compliance required from August 2026 and penalties up to 30,000,000 euros or 6% of global turnover for serious violations. GDPR Article 22 requires that candidates not be subject to solely automated decisions with significant effects without a documented human review option. In the United States, New York City and Colorado require mandatory auditing of AI hiring systems. Across jurisdictions, the baseline requirements are documented bias testing, audit logs, human oversight mechanisms, and candidate disclosure that AI is being used in the screening process.