Artificial intelligence is reshaping every stage of the employment relationship. This creates both tremendous opportunities and legal risk. Drawing on case law, regulatory guidance and real-world examples, this four-part series discusses different aspects of the HR world where AI is already creating legal exposure for employers and what HR professionals can do right now to get ahead of it. The best place to start is the recruiting and hiring stage.
Employers nationally have quickly adopted AI products throughout the hiring process. It is easy to see why: A recruiter facing thousands of applications for a single posting cannot meaningfully review every single one. AI tools can sort, rank, and surface the strongest candidates in seconds, all before any recruiter or human resources representative reviews an application.
The same technology can also widen a candidate pool. AI can surface qualified applicants who may have been missed by a keyword search and reduce the inconsistency created when different hiring managers apply their own unwritten standards to different stacks of resumes. AI does not simply make hiring faster; it can make hiring more consistent and predictable. On its face, this can look fairer than a purely human process riddled with its own unexamined biases.
AI is, however, only as smart as the rules a human builds into it. The uncomfortable truth about AI and its integration within the employment relationship is that employment law risk does not generally flow from a rogue algorithm taking a step an employer never intended. Instead, the risk comes from a hiring team using an AI system as part of its decision-making process but not completely understanding it.
AI hiring: A popular litigation target
AI hiring tools create the opportunity for expensive and damaging class actions. One such high-profile case, Mobley v. Workday, Inc., tests whether HR technology vendor Workday violated state and federal anti-discrimination law by creating an AI-powered applicant screening tool that discriminated on the basis of age and disability. The case remains in discovery after the Court granted preliminary collective certification, paving the way for potentially millions to join the collective action.
Similarly, in Kistler v. Eightfold AI Inc., another HR vendor is currently facing suit for allegedly acting as an unregistered consumer reporting agency, scraping data on more than 1 billion workers, and scoring applicants on a 0-to-5 “likelihood of success” scale, without making the requisite disclosures under the Fair Credit Reporting Act.
While the third-party vendor may create these screening tools, employers utilizing biased tools may also face liability. In Harper v. Sirius XM Radio, LLC, an employer faces multiple discrimination claims premised on its use of an AI system that allegedly evaluated candidates using data points functioning as unlawful proxies for race. If the Harper case teaches employers anything, it should be that using third-party vendor tools will not insulate them from litigation and, ultimately, liability.
AI does not eliminate accommodation needs
Overreliance on AI tools without human involvement creates particular exposure to claims of disability discrimination. This reality led the DOJ and EEOC to issue guidance in 2022 on how employers’ use of hiring technologies may violate the ADA. For example, a hiring tool built to predict “who will be a good employee” by comparing candidates to current successful staff can exclude people with disabilities, simply because these individuals were underrepresented in the “good employee” comparison pool.
Likewise, tools like facial or voice analysis assessments can screen out applicants with certain disabilities like autism or speech impairments without providing them the opportunity to request an accommodation. Information about an assessment and how to request an accommodation should be visible before a candidate starts the evaluation. A human-reviewed alternative can serve as a reasonable accommodation, allowing employers to evaluate these candidates fairly. Without a meaningful accommodation process, employers risk unnecessarily, and potentially unlawfully, excluding disabled but qualified applicants.
The shifting regulatory framework
Governmental actors have also addressed the rising use of AI in hiring. As referenced earlier, Biden-era DOJ-EEOC guidance highlighted how hiring technology may violate the ADA. A subsequent executive order has since withdrawn this guidance, but a shift in administration may also reignite federal interest in the topic.
In recent years, cities and states have acted where the federal government has not. These efforts have focused on disclosing AI use, mandating internal audits and ensuring meaningful human involvement in decisions. For example, New York City requires a bias audit before an employer can use an “automated employment decision tool,” plus public disclosure and candidate notice. Illinois is taking multiple actions, one for AI video interview analysis and another for discriminatory impact generally. California, Colorado, New Jersey and Oregon are among the states taking legislative or administrative action in light of the federal absence. Short of significant congressional action, employers should prepare to face a patchwork of state and local regulation on this topic.
Candidates are adapting too
The AI arms race runs in both directions. Candidates increasingly prepare their submissions with AI screeners in mind. For example, many applicants are adding hidden white-on-white text meant to manipulate AI screening tools. While invisible to the human eye, these so-called prompt injections surreptitiously direct the AI screener to give the candidate a favorable ranking regardless of the content of the resume. A recent Duke University study discovered that 1% of roughly 200,000 resumes within the dataset contained prompt injections.
Candidates are also using AI to draft everything from cover letters to coding samples. Employers seeking to discover and exclude AI-drafted materials should do so cautiously. Technologies that attempt to catch AI-drafted materials are flawed and may create their own exposure. Stanford research found leading AI detectors misclassified more than 60% of essays written by non-native English speakers as AI-generated, because the parameters reward linguistic sophistication that native speakers produce more easily. Future tools may close that gap, but today’s tools are unreliable enough that using them may cause more harm than good.
Steps to reduce risk and improve outcomes
Given the rise of litigation around AI-powered hiring, employers should inventory every tool that screens, ranks or recommends. Yes, that includes the ones you’ve used for years but never labeled as “AI.” From there, leaders should ask difficult questions of their HR technology teams, both internal and external, to understand the parameters of their AI-powered decision-making processes.
Once leadership understands the parameters built into the system, employers should ensure that the system works as desired. Employers benefit from building tools that periodically audit the system’s results to identify potential bias. Given the prompt injection risk, employers should safeguard their systems by stripping formatting before any application document goes through an AI reviewer.
Beyond the internal technological solution, employers also benefit from disclosing the use of AI after the initial screening of application materials. Employers that inform applicants what AI tools are in use and how to request an accommodation, if needed, allow qualified candidates with disabilities to fully and meaningfully participate in the hiring process.
Taking these steps will reduce litigation risk and prepare employers for the increasingly complex patchwork of state and local regulation. Moreover, these steps will both improve efficiency and expand the number of qualified candidates in an employer’s applicant pool.
The post Recruiting and hiring: Where AI delegation meets exposure and exclusion appeared first on HR Executive.
Read MoreHR Executive