Policy

Policy Article 4 Reaches the Interviewers Who Operate the Tool Article 4 has applied since 2 February 2025 and lands on whoever operates AI on the employer's behalf. Which interviewers that covers, and what discharging it looks like.

Policy When the Tool You Banned Is Someone's Assistive Technology A candidate who uses AI to manage a disability is raising a question about what your round measures. What to ask, what never to ask, and what to offer instead.

Policy Put the AI Expectation in the Role, Not the Offer Letter An offer letter carries title, pay, start date and at-will language. Put the AI expectation in the written role scope and the first performance plan instead.

Policy When an AI Interview Asks Something It Should Not, Keep the Record A scripted question can't be redirected. Where the ADA draws the pre-offer line, why a graduation year is the same question, and what to save before the tab closes.

Policy When the Transcript Gets It Wrong, the Score Reads the Mistake Where the AI scores the transcript, a recognition error changes what gets scored. What the vendors publish, how big the measured gap is, and what to send the recruiter.

Policy Check What They Decided, Not Whose Project It Was Almost all workplace AI work is collaborative and the interview asked for the first person. Before the offer moves, ask which decisions inside the project were theirs.

Policy When an AI Recruiter Has to Say What It Is From 2 August 2026 the EU AI Act makes systems that talk to you say so. US rules attach to analysis, not the voice. What to ask, and how to keep the answer.

Policy An AI Requirement Answers to the ADEA, Not to the Four-Fifths Rule An AI requirement that leaves out everyone over fifty is an ADEA question, not a Title VII one. The selection-rate math to run, and the defense an employer carries.

Policy Your AI Requirement Screens Before the First Application Arrives Selection-rate math needs applicants, and a posting requirement removes people before they become one. Test the requirement against the task, then vary the wording.

Policy Trade Secrecy Is a Limit on the Answer, Not a Reason to Give None What the EU's top court said an explanation must contain, why a trade secret claim routes the dispute to a regulator instead of ending it, and the dated US rights.

Policy Don't Ask a Candidate to Open Their Own AI History A personal AI account holds a former employer's material and the candidate's own health and money questions. Supply the session you want to watch instead.

Policy Ask for a Human Rater Before the Window Closes One major platform documents a manual-rating path for candidates who decline AI scoring. What to ask for, who decides it, and why timing matters more than wording.

Policy A Biometric Consent Screen Can Cover Three Different Things One checkbox before an AI interview can cover a recording, a face or voice template, or automated analysis. Only one of the three is biometric in law.

Policy A Candidate's Agent Needs the Review You Give a Vendor What has to be true before a candidate's agent gets credentials: scoped access, synthetic data, no production route, an expiry set at creation, a logging notice.

Policy What the Position Statement Has to Show About the AI Round A charge can bring a position statement on about a 30-day clock and a request for information. What the requisition file has to already hold, and who reads the answer.

Policy The AI Productivity Studies Randomized a Tool, Not a Person The AI productivity trials randomized a tool, never a person, so they carry no validity evidence. What legal can rely on instead, in the Uniform Guidelines' own terms.

Policy Why an Employer Can Interview You With AI and Ban Yours One side picks the tool, the other changes what it reads. That is fair. Silence is not: in a vendor survey, 70% of US candidates said nobody clearly told them up front.

Policy Send the Preservation Request Before the Complaint Ask the employer in writing to hold the recording, transcript and scores before you raise a problem. A deletion request destroys the same file you would need.

Policy Where Scoring a Face or a Voice Is Actually Barred Emotion inference at work has been prohibited in the EU since February 2025. In the United States there is no general ban, and only two states require consent.

Policy Who Scores the Take-Home Decides Which Rules It Falls Under A model can grade take-home submissions. Scoring by machine adds audit, notice and explanation duties in some jurisdictions, and the score itself moves with input order.

Policy What the Model Was Trained On Decides What It Rewards Some hiring models are fitted to a client's own high performers, others to how a rating panel saw a video. What the training target changes, and what to ask.

Policy Write the Retake Rule Before the First Person Asks Three defensible retake rules for an AI exercise, why one retake for everyone usually wins, and what changes the moment the request names a medical condition.

Policy An AI Summary Changes What Your File Has to Show New York City's rule reaches a score, tag or recommendation, not a purely narrative summary. What changes is the file: label it, keep it, record what came first.

Policy Write What You Heard, Never Why You Think It Happened The words that hurt later are the ones explaining why. Swap every guess about a candidate's process for the question asked, the words used, and the requirement.

Policy Your Sourcing Tool Ranks People Who Have Not Applied A tool that ranks profiles before anyone applies sits outside New York City's rule and outside every selection rate you compute. What still reaches it, and what to keep.

Policy The UK Replaced Article 22: What to Ask For Now The Data (Use and Access) Act 2025 replaced GDPR Article 22 in the UK. What Articles 22B and 22C entitle a candidate to ask for, dated, and how to word the request.

Policy Your AI Exercise Became a Test the Day It Moved a Decision Federal selection law draws the line at use, not paperwork. An informal, unscored AI exercise is a selection procedure the first time it changes who advances.

Policy Your Interview Data Request Belongs With the Employer, Not the Vendor HireVue's own pages say the employer decides on your interview data. Where to send the request, what to name in it, and which law puts a clock on it.

Policy Your Rights Before an AI Interview, by State and Date Federal law gives you no general right to know AI scored your interview. A few states already do, each with its own effective date. Here is the table.

Policy You Cannot Compare the Model to Recruiters You Never Measured No employer holds a measurement of its own screeners, so the comparison has no baseline. The evidence on both sides, and the question that can be answered instead.

Policy What You're Agreeing To in a Candidate AI-Use Policy A candidate AI-use agreement usually covers five clauses: assistance, stages, recording, attestation, and consequence. Here is what's ordinary in each.

Policy Amazon's Tool Learned Its Bias From Amazon's Own Hiring Reuters reported an internal experiment, abandoned before it gated anyone. The detail that matters is what it was trained on, and why one deleted word settles nothing.

Policy The Application Asks If You Used AI: Answer It Straight Answer the AI-use box accurately. It is a low-stakes question with a high-stakes wrong answer, since application accuracy is usually part of the offer.

Policy Extra Rights You Get Applying to a European Employer GDPR gives an EU applicant a right no US law matches: human intervention against a solely automated hiring decision. Here is the request to send, and its limits.

Policy Ask on a Separate Voluntary Form the Decision-Makers Never See Voluntary, separated from the application, invisible to anyone who evaluates. What the EEOC actually says, where the data has to sit, and how to count declines.

Policy How to Ask for a Human Review After an Automated Rejection Most US employers owe no general right to a second look after an automated rejection. Here are the real signals one exists anyway, and how to ask for it.

Policy Conditional Offer First, Then Two Notices Before You Reject The order: conditional offer, standalone disclosure and written authorization, then a pre-adverse notice with the report and a real interval before the final one.

Policy A Bias Audit Scores the Tool You Bought, Not the Way You Use It A bias audit computes selection rates and impact ratios by category, on a chosen data set, on one date. Here is what it never covers, and who owns that.

Policy What an Employer Can and Can't See About Your AI Use Nothing in a personal AI account on your own device is visible to an employer. Inside an environment they control, a tool captures what it was built to capture.

Policy Can a Company Train Its AI on Your Application? Yes, if the privacy notice separately discloses it and a lawful basis supports it. Purpose limitation blocks silent repurposing; California applicants hold real rights.

Policy A Trial Period You Direct Is Employment, Whatever the Contract Says Use a contractor for a defined deliverable from an independent practice. Hire an employee for your work, your way, your schedule, which is what a trial month is.

Policy When Saying You Used AI Helps, and When It's Noise Disclose where AI's contribution changes what the document proves, and stay quiet everywhere else. A resume and a portfolio project are not the same case.

Policy Does a Human Ever Watch Your AI-Scored Interview? Not the way you'd think: an AI-scored interview ranks or shortlists rather than rejects outright, and a person enters at the shortlist to decide.

Policy Where You're Entitled to Know AI Screened Your Application No federal law makes an employer tell you AI screened your application. A short list of states and cities already requires notice. Check whether yours is on it.

Policy What You Hand Over When a Bot Applies From Your Account Auto-apply tools want a login first. The credential grants more than the task, and LinkedIn's terms ban the automation outright. Say yes narrowly, if at all.

Policy Ask What the Job Requires, Not What the Person Is Federal law gives a job-relatedness test, not a banned list, with the ADA as the sharp exception. Most barred questions have a substitute that gets what the job needs.

Policy Using AI to Apply Isn't Cheating. Breaking a Stated Rule Is. AI on an application isn't dishonest by itself. It becomes dishonest when you claim work you didn't do after a rule told you not to. Three cases, one test.

Policy Off-List Reference Calls Produce Evidence the Candidate Cannot Answer A backchannel call is mostly lawful and structurally one-sided: it puts a judgment in the file the candidate never sees. One rule resolves nearly every case.

Policy Can an Offer Be Rescinded Over AI Use on the Application? Rarely for AI use alone. The real risk is a false attestation, a wrong disclosure answer, or a claimed credential that doesn't exist. Here's the honest range.

Policy An Outcome Gap Names a Stage to Inspect, Not a Verdict One group passing at a lower rate is evidence about a stage, not a finding about the process. Here is what to hold constant before reading a ratio at all.

Policy A Probation Period Is a Deadline for the Manager, Not a Legal Shield What a probationary period does in US law: almost nothing federally, everything in Montana, one useful thing everywhere. Keep the date, check the label with counsel.

Policy Run Both: The Ratio Sizes the Gap, the Test Says Whether It Is Real The ratio is an effect size and the test is a detectability check. They fail in opposite directions at opposite volumes, so run both per gate and read the disagreement.

Policy How to Read a Published Bias Audit Before You Apply A posted bias audit reports two numbers, not a verdict on fairness to you. What the ratio measures, what the 2% carve-out hides, and what a clean audit does not prove.

Policy Assume Anything You Send a Candidate Goes Into a Model Settle sensitivity before confidentiality. Assume the brief and its attachment get pasted into a chat window, then build a synthetic dataset with the same shape.

Policy Can You Record Your Own Interview or Bring a Notetaker? Federal law lets one party to a call record it. Some states need everyone's consent. Here is the map, the ask that costs nothing, and what the employer owes you too.

Policy Where to Report an AI Hiring Decision You Think Discriminated Four doors exist and one has a hard clock. The EEOC deadline, the state and city options, what to preserve now, and the honest odds before you spend months on it.

Policy How to Ask for an Accommodation on an AI Interview Name the barrier, not the diagnosis, and ask before you're scheduled: what to request, who to ask, and what the ADA actually requires from the employer.

Policy Vary the Data, Never the Task or the Bar Everyone at the same stage gets the same task, materials, stated time and bar. Vary the inputs to defeat sharing; varying the task creates a second procedure.

Policy Fix the Questions and the Anchors; the Follow-Ups Can Differ Every candidate faces the same planned questions and the same written anchors. Probes may differ in wording. What must never differ is who gets probed at all.

Policy Whose AI Hiring Law Covers You When You Apply Out of State The trigger for most AI hiring laws is your address or the job's location, not where the company is headquartered. A table of what triggers what, and when.

Policy Do You Allow a Candidate's AI Agent to Apply and Schedule? Allow an agent to fill forms, book slots and answer scheduling email. Require the person for the assessment, the interview and every signature, and say so in the invite.

Policy How Do You Run an Adverse Impact Audit When the Vendor Holds the Data? Run the four-fifths ratio on your own ATS data. Under 0.80 is evidence of adverse impact, not proof, and occupation-specific validity evidence decides if the tool stays.

Policy The Four-Fifths Rule Is a Trigger for Scrutiny, Not the Legal Threshold Adverse impact is a substantially different rate of selection. The four-fifths rule flags ratios under 0.8 for enforcement attention, and it is not the legal test.

Policy What Does HR Have to Build in 90 Days After an AI-First Mandate? Three things must exist inside 90 days: an inventory of AI uses touching employment decisions, notice and records where a statute reaches, and one named owner.

Policy Keep the Output, the Version, the List You Saw, and the Reason Federal recordkeeping runs one year, California's FEHA rules four. The harder half is the object list: the output, the version, the list you saw, and the reason.

Policy AI Hiring Laws Regulate Your Tools, Not the Candidate's NYC Local Law 144, Illinois, Texas and the EU AI Act all bind the employer's tool. The exposure runs the other way: through the detector you would buy to enforce a rule.

Policy What Your AI Notice Must Say, and When Each Jurisdiction Wants It New York City wants ten business days of warning. Illinois wants notice from January 1, 2026 and consent before AI reads a video interview. California wants records.

Policy Set the AI Rule Stage by Stage, and Let the Job Decide Each One Ask one question of each stage: with an assistant open, does it still measure what it exists to measure? The rule follows the answer, stage by stage.

Policy Vendor Indemnification Does Not Move Employer Liability. It Moves Who Pays. An indemnity is a contract about cost. It does not decide who a candidate can name, and it cannot shift a duty a discrimination statute puts on the employer.

Policy Should Interns Be Allowed to Use AI on the Take-Home, and How Do You Grade It? Allow it, require the record, and grade four decisions the model can't make: scope set, claim checked, work kept, output refused. What earns credit varies by field.

Policy Should You Allow AI on the Take-Home Assignment? Match the rule to the job's real working conditions: a three-line policy, the exact wording for the brief, and what to grade once AI is allowed.

Policy Ask What They Did With It, Not Whether They Used It Ask candidates what they asked the assistant for and what they changed or threw away, with a printed commitment that an honest answer alone never ends the process.

Policy Your ATS Shipped AI You Never Bought A release note can hand an employer an automated decision tool and a candidate-notice duty with no purchase to attach a review to. Off is the defensible starting state.

Policy A Blanket AI Ban Only Works on the Candidates Who Obey It A ban is a request with a penalty nobody can impose, because the checking does not work. What it reliably does is sort the pipeline by who follows instructions.

Policy The Five Clauses a Candidate AI-Use Policy Actually Needs The stage, what is allowed there in verbs, what you ask the candidate to tell you, what you do with the answer, and a consequence you can evidence.

Policy Does Showing Candidates Their Report Raise or Lower Legal Exposure? Parity lowers exposure when a report records acts and raises it when a report infers traits. The send test, what to cut before release, where disclosure is already owed.

Policy Can You Stop Candidates From Recording Interviews for AI? You can require disclosure, but nothing tells you a phone is running. The rule that holds is symmetric: what you record, what they record, the scope, how long it's kept.

Policy How Do You Defend an AI-Skills Assessment to Legal and to Candidates? Two artifacts carry the defense: a job-relatedness file rebuilt per role, and the candidate's own copy of the result. Validity evidence does not transport between jobs.

Policy Will an AI-Skills Assessment Hold Up if a Candidate Challenges It? Under Title VII you must show the assessment is job related. Six records built before the first invite carry that; a generic AI score carries none of it.

Policy Asking Candidates to Disclose AI Puts Your Own Stack on the Table No US rule pairs the two duties: notice attaches to the tools an employer deploys. Practically they are one paragraph, because one-way transparency is remembered.

Policy Disclosure, Detection, or Observation: Which AI-Use Policy Holds Up? Detection can't carry a decision. Honor-system disclosure works only where AI isn't part of the job. Observation is the option that produces evidence.

Policy One US Jurisdiction Requires a Bias Audit. Testing Matters in the Rest Anyway. Only New York City mandates an annual independent bias audit. What that rule actually covers, and why California now weighs anti-bias testing as evidence in a claim.

Policy The EEOC Pulled Its AI Guidance. Title VII Did Not Change. The EEOC's AI technical assistance came off its site in January 2025. Title VII, the ADEA and the ADA still apply, and the live 2026 case is a private one.

Policy Your Employee AI Policy Does Not Reach the Candidate An acceptable-use policy binds through employment, names tools an applicant cannot reach, and controls a different risk. Keep two documents that share one sentence.

Policy Enforce the AI Rule by Design, Not by Watching Three changes make an AI rule self-enforcing: ask for choices instead of artifacts, use material no model has seen, and talk about the work afterwards.

Policy What's a Safe Answer When the EPLI Form Asks About AI in Hiring? Every tool that touches a hiring decision gets a dated line: what it outputs, which stage it runs at, who reads it before a rejection, and whether an audit exists.

Policy The EU AI Act Reaches US Employers, and the Hiring Deadline Moved to 2027 Regulation (EU) 2024/1689 follows the output, not the office. High-risk hiring duties moved to 2 December 2027, but the prohibitions have bound since February 2025.

Policy What Do You Owe a Candidate the AI Screened Out? Notice, an accurate account of what the tool evaluated, and a route to a human. From 2027 Colorado adds the tool's role in plain language, correction and human review.

Policy What Makes an AI Requirement in a Posting Defensible In the US you can require AI experience. What binds is the job-relatedness showing behind it: the tasks it came from, a written bar, and evidence applied evenly.

Policy Can You Legally Ask Candidates How They Use AI? Asking how a candidate works with AI is a lawful job-related question. Exposure starts when a tool scores the answer or the question invites disability disclosure.

Policy Hiring for AI You Have Not Approved Internally If the tool is blocked at the proxy, write the requirement in judgment terms and name the approval date, or the hire spends a quarter working around it.

Policy Can You Rescind an Offer After the Take-Home Broke Your AI Rule? At-will means you almost certainly can. It holds up only if the AI rule was in the brief before the task and every candidate who got that brief is held to it.

Policy No US Law Gives Candidates a Right to a Human Reviewer. Europe Does. No US statute in force gives a candidate a right to human review. GDPR Article 22 does, it follows EU-based candidates, and Colorado adds a request right in 2027.

Policy Should You Let Candidates Use AI During the Interview, or Ban It? Permit exactly what the job permits. Banning AI for a role that runs on it tests a skill nobody uses; allowing it where the work forbids it tests the wrong thing.

Policy What's the Smallest Hiring Process a Two-Person Company Can Defend? Four artifacts a two-founder team writes in an afternoon, the employee counts that decide which laws bind you, and the short record the Uniform Guidelines allow.

Policy What Belongs in an AI Hiring Policy, and Who Has to Sign Off? The eight sections a hiring AI policy needs, who owns each, and the four signatures that make it enforceable. Start with permitted AI use by job family.

Policy Three Tests Decide Whether Your Tool Is an AEDT Under Local Law 144 The NYC rules write the trigger down: a simplified output, plus one of three ways a human uses it. Reviewing every candidate clears only the first of the three.

Policy Five US AI Hiring Laws Bind You Today, and Colorado Is Not One As of August 2026, five US regimes impose real duties on employers using AI in hiring: NYC Local Law 144, two Illinois statutes, Maryland, and California's FEHA rules.

Policy The AI Rule Belongs to Whoever Reads the Work Legal owns what you may ask and keep. HR owns consistency across requisitions. The person reading the output owns the rule for their stage, and the role gets named.

Policy What Do You Have to Tell a Candidate Before Recording Their Screen, and How Long Can You Keep It? Say what is captured, why, who watches and how long you keep it before the candidate starts. Record the assessment surface only, and treat a decline as a normal path.

Everything on policy