Photo by: Raphaël Biscaldi
As artificial intelligence becomes more deeply embedded in recruitment, employers face a growing question over how much candidates should know about the systems evaluating them.
For Sebastian Scott, CEO of Clera, an AI talent agent initiating better hiring conversations, he says transparency is essential, but avoiding AI is not the answer.
For many roles, employers now receive thousands of applications, making it increasingly difficult for a person to properly review every profile. Without technology, Scott argues, many candidates may never be meaningfully considered.
“Used well, AI can therefore be in candidates’ interests because it gives more people a genuine chance to be seen,” he said. “The real breakthrough isn’t automation. It’s attention.”
But the challenge, he said, is explaining how an AI system reaches its conclusions. With complex models, a simple explanation can risk being misleading. Candidates do not necessarily need access to source code, but employers should explain what information was considered, what the system was designed to assess and how its output influenced the decision.
That transparency also needs to be paired with accountability. Systems should be auditable and regularly tested for consistency and bias, while candidates should have meaningful opportunities for human review. They should also be able to correct inaccurate information and challenge an outcome.
This becomes especially important when an AI system is used at an early stage of the hiring process, where a decision can determine whether a candidate ever reaches a human recruiter. A candidate who is screened out may have no way of knowing whether the decision was based on their actual qualifications, an inaccurate data point that the employer did not realize the system was using. Transparency gives candidates a clearer understanding of what happened and gives employers a way to identify problems before they become systemic.
Bias is particularly difficult to address because AI systems learn from data shaped by previous human decisions. But Scott said bias can enter the process in other ways, including through the signals a system is asked to consider and the definition of success it is designed to predict.
A university, previous employer, career gap or even the wording of a résumé can become a proxy for characteristics that an employer never intended to measure. The key question, he continued to say, is whether a particular signal genuinely predicts someone’s ability to perform a job or simply reflects who was hired in the past.
Scott also sees AI as an opportunity to address some of the hidden biases already present in human hiring. Recruiters reviewing thousands of applications inevitably rely on shortcuts, such as familiarity with certain companies or universities, or an intuitive idea of what a strong candidate looks like. Those judgments can be inconsistent and difficult to examine.
Technology, he argues, can instead allow employers to consider more people against explicit, job-relevant criteria and test whether those criteria produce biased outcomes. But that requires continuous oversight rather than assuming an algorithm is inherently objective.
Scott is also concerned about the possibility of what he describes as an “algorithmically blacklisted” candidate. Such a blacklist would not need to exist formally. If multiple employers use similar models, data sources or proxies, candidates could repeatedly be excluded for the same underlying reason.
A career gap, non-traditional education or previous job title could become a negative signal across the employment market. The result could be systemic exclusion, with individual employers believing they are making independent decisions while reproducing the same bias.
The concern becomes greater as AI hiring tools become more widespread. If the same assumptions are embedded across multiple systems, a candidate could face the same barrier repeatedly without ever knowing why. What appears to be a series of separate hiring decisions could effectively become a shared filtering mechanism, narrowing opportunities for certain candidates across an entire industry.
Scott argues that assessments should therefore remain specific to a particular role and moment in time. A rejection or low score should not become a permanent label that follows someone between employers. Candidates should be able to correct inaccurate information and access meaningful human review.
Ultimately, he places responsibility on both employers and technology providers.
“Accountability should not be treated as the opposite of innovation,” Scott said. “It is what allows this technology to earn trust and deliver its benefits.”
For employers, that means treating AI as a decision-support tool rather than an unquestionable authority. Technology providers, meanwhile, have a responsibility to make their systems understandable enough for employers to evaluate, monitor and challenge their outputs. Neither side can assume that the other is solely responsible for the consequences of an automated hiring decision.
Finally, for Scott, the goal is not to automate existing hiring preferences. It is to examine them, test them and improve them while using technology to give more candidates a genuine opportunity to be considered.
For more business and technology coverage, visit Entreprenerd.