[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"footer-data":3,"$f1rh8n39iphevj":42},{"countries":4,"agencies":26},[5,10,14,18,22],{"id":6,"name":7,"agencyCount":8,"avgAgencySize":9},"brazil","Brazil",989,0,{"id":11,"name":12,"agencyCount":13,"avgAgencySize":9},"liechtenstein","Liechtenstein",21,{"id":15,"name":16,"agencyCount":17,"avgAgencySize":9},"british-virgin-islands","British Virgin Islands",2,{"id":19,"name":20,"agencyCount":21,"avgAgencySize":9},"dominican-republic","Dominican Republic",70,{"id":23,"name":24,"agencyCount":25,"avgAgencySize":9},"madagascar","Madagascar",52,[27,30,33,36,39],{"_id":28,"name":29,"id":28},"69301514b3404633fe0f16db","Horton International",{"_id":31,"name":32,"id":31},"69301518b3404633fe0f1721","buro Nexus",{"_id":34,"name":35,"id":34},"69301517b3404633fe0f1710","Catenon",{"_id":37,"name":38,"id":37},"69301519b3404633fe0f1731","Allegis Group",{"_id":40,"name":41,"id":40},"69301516b3404633fe0f1701","Robert Half",{"article":43},{"slug":44,"title":45,"html":46,"topic":47,"coverUrl":50,"coverAlt":51,"publishedAt":52,"agency":53,"related":54,"comments":63,"alternates":64},"ai-screening-does-not-move-the-blame","AI screening does not move the blame for a bad hire","\u003Cp>An employer that hands its first screen to software, directly or through an agency, often assumes it has handed over part of the risk as well. It has not. When a tool filters out the wrong people, the people asking questions afterwards will put them to the company that made the hire, and to the agency that ran the search on its behalf.\u003C\u002Fp>\n\u003Cp>This piece argues that the blame for an AI screening mistake never moves to the algorithm or to the vendor that sold it. It then sets out what people in recruiting are asking for as standards, and the questions an employer can put to any agency that screens with AI.\u003C\u002Fp>\n\u003Ch2 id=\"the-decision-still-has-an-owner\">The decision still has an owner\u003C\u002Fh2>\n\u003Cp>Writing on ERE in August 2026, Raghav Singh put the principle plainly. The point is simple, and it is easy to forget once a tool is running quietly inside an applicant tracking system.\u003C\u002Fp>\n\u003Cblockquote>employers cannot outsource accountability simply because an AI system participated in a decision.\u003C\u002Fblockquote>\n\u003Cp>A screening tool does not hire anyone. It ranks, scores, or drops applications, and a person or a process the employer set up acts on the result. If the result is wrong, the question is who chose the tool, who set the criteria, and who checked the output. Those are management decisions, and they belong to whoever owns the hiring process.\u003C\u002Fp>\n\u003Cp>The same logic applies when an agency does the screening. The agency chose and configured the tool, so it answers for the shortlist it sends. The employer chose the agency and accepted the shortlist, so it answers for the hire. Neither one can point at the software and step aside.\u003C\u002Fp>\n\u003Ch2 id=\"the-rules-already-point-at-the-user\">The rules already point at the user\u003C\u002Fh2>\n\u003Cp>This is not only an ethical argument. Several public rules and cases already put responsibility on the organisation that uses the tool, not only on the company that built it.\u003C\u002Fp>\n\u003Cp>In New York City, \u003Ca href=\"https:\u002F\u002Fwww.nyc.gov\u002Fsite\u002Fdca\u002Fabout\u002Fautomated-employment-decision-tools.page\">Local Law 144 of 2021\u003C\u002Fa> prohibits employers and employment agencies from using an automated employment decision tool unless it has had a bias audit within one year of use, a summary of that audit is public, and candidates have been notified. The city's consumer and worker protection department began enforcing it on 5 July 2023. The law names employment agencies alongside employers, which matters for anyone who outsources the first screen.\u003C\u002Fp>\n\u003Cp>In the European Union, the AI Act lists AI systems used for recruitment or selection, including tools that \u003Ca href=\"https:\u002F\u002Fartificialintelligenceact.eu\u002Fannex\u002F3\u002F\">analyse and filter job applications and evaluate candidates\u003C\u002Fa>, as high-risk. Its \u003Ca href=\"https:\u002F\u002Fartificialintelligenceact.eu\u002Farticle\u002F26\u002F\">article on deployers\u003C\u002Fa> asks the organisation using such a system to assign human oversight to people with the competence and authority to exercise it, to keep the system's logs, and to tell people that such a system is being used on them. The same page dates those obligations for Annex III systems to 2 December 2027.\u003C\u002Fp>\n\u003Cp>In the United States, a lawsuit against Workday shows where disputes can land. In May 2025 a federal judge in California \u003Ca href=\"https:\u002F\u002Fwww.hrdive.com\u002Fnews\u002Fworkday-ai-bias-lawsuit-class-collective-action\u002F748518\u002F\">preliminarily certified a collective action\u003C\u002Fa> for applicants aged 40 and over who say an AI recommendation system screened them out. Workday denies the claims, and the court has not decided whether they have merit. The case is still a reminder that the employers using such a platform cannot assume the matter stays between the plaintiffs and the vendor.\u003C\u002Fp>\n\u003Ch2 id=\"a-vendor-audit-is-not-a-standard\">A vendor audit is not a standard\u003C\u002Fh2>\n\u003Cp>Many tools now arrive with a fairness audit attached. In a September 2026 ERE column, David Manaster argued that these audits work mostly as reassurance for buyers, and that they differ in scope, method, and what counts as a pass because no shared standard exists. An employer reading a vendor's audit is reading the vendor's own definition of fair.\u003C\u002Fp>\n\u003Cp>Manaster proposed a short set of lines that any AI screening should respect. In plain terms, they come to four points.\u003C\u002Fp>\n\u003Cul>\n\u003Cli>A candidate is declined automatically only on a criterion the employer can state in advance, in plain language, as a requirement of the job.\u003C\u002Fli>\n\u003Cli>No one is declined on an attribute the employer never specified and the system inferred on its own.\u003C\u002Fli>\n\u003Cli>Candidates are told before AI evaluates them, and that notice is built into the product rather than left as a setting.\u003C\u002Fli>\n\u003Cli>Anyone screened out by software can ask for a human to look again.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>He put the last point in words worth keeping.\u003C\u002Fp>\n\u003Cblockquote>Any candidate screened out by software can request human reconsideration and actually receive it.\u003C\u002Fblockquote>\n\u003Cp>None of these lines needs new technology. They need someone on the hiring side to write down the criteria, read what the tool does with them, and stay answerable for the result. That is the accountability argument in practical form.\u003C\u002Fp>\n\u003Ch2 id=\"what-to-ask-an-agency-that-screens-with-ai\">What to ask an agency that screens with AI\u003C\u002Fh2>\n\u003Cp>An employer briefing an agency can turn this into a short set of questions before the search starts. The answers show whether the agency treats its tools as something it owns or as something it simply switched on.\u003C\u002Fp>\n\u003Col>\n\u003Cli>Which tools touch our candidates, at which step, and who at the agency approved each one?\u003C\u002Fli>\n\u003Cli>Which criteria can reject a candidate without a person reading the application, and who wrote them?\u003C\u002Fli>\n\u003Cli>When was each tool last checked for bias, by whom, and can we read the summary?\u003C\u002Fli>\n\u003Cli>Are candidates told that AI is part of the screen, and when?\u003C\u002Fli>\n\u003Cli>Can a rejected candidate ask for a human review, and who does it?\u003C\u002Fli>\n\u003Cli>What record do you keep of what the tool recommended and what a recruiter decided?\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>An agency that can answer these quickly has probably thought about them already. An agency that cannot is asking the employer to share a risk that nobody is managing.\u003C\u002Fp>\n\u003Ch2 id=\"keep-the-blame-where-the-decision-is\">Keep the blame where the decision is\u003C\u002Fh2>\n\u003Cp>AI screening can save a hiring team a great deal of time, and nothing here argues against using it. The argument is narrower: the tool changes how a shortlist is made, not who answers for it. The employer and the agency that use the tool own its mistakes in the same way they own a recruiter's mistakes.\u003C\u002Fp>\n\u003Cp>The practical step is to put that ownership in writing before the first application arrives. Name the person who approves each tool, write the knockout criteria in plain language, ask for the audit summary, and agree how a rejected candidate can reach a human. When something goes wrong, the company that did this has an answer ready.\u003C\u002Fp>",{"slug":48,"name":49},"hiring","Hiring","\u002Fuploads\u002Farticles\u002F6ad718639391cec11c80fff3\u002Fcover.jpg","A desk seen from above with a laptop showing a grid of candidate profiles, a stack of CVs, and one CV set apart under a pen.","2026-10-20T07:39:13.153Z",null,[55],{"slug":56,"title":57,"excerpt":58,"topic":59,"coverUrl":60,"publishedAt":61,"publishedLabel":62},"candidate-fraud-remote-hiring","How agencies can spot candidate fraud in remote hiring","What candidate fraud looks like in remote hiring Candidate fraud covers a wide range, from a CV that stretches the truth to a person who is not who they claim to be. For an agency placing remote tech…",{"slug":48,"name":49},"\u002Fuploads\u002Farticles\u002F6ac69c49c3683e93324a7f24\u002Fcover.jpg","2026-10-07T19:23:53.853Z","7 Oct 2026",[],[65,67,70,73],{"locale":66,"slug":44},"en",{"locale":68,"slug":69},"hi","ai-screening-se-galat-bharti-ki-zimmedari-nahin-hatti",{"locale":71,"slug":72},"de","ki-screening-verschiebt-die-verantwortung-nicht",{"locale":74,"slug":75},"es","la-evaluacion-con-ia-no-traslada-la-culpa"]