Matt Chambers, Founder and CEO of unified AI recruiting platform Loxo, says evidence of unpredictable bias and growing legal exposure is giving executive search firms a clearer picture of where AI can go wrong.
As courts and regulators put greater scrutiny on AI’s role in employment decisions, including an ongoing federal lawsuit testing whether a technology vendor can share responsibility for alleged discrimination produced by its screening tools, executive search firms are grappling with a complex question: How can AI tools be used to recruit better candidates?
It’s a question that may still be premature. Studies and surveys have shown that artificial intelligence can develop hiring biases that humans never programmed into it, even while employers are finding that the technology can surface candidates traditional screening might overlook.
There isn’t yet a reliable playbook for succeeding with AI in executive search, but there is an increasingly visible collection of ways to fail with it, and understanding those failures, including where human judgment can prevent them, is where firms should start.
The first phase of AI adoption in recruiting focused heavily on efficiency. Firms experimented with using the technology to build longlists, match candidates, draft communications and clear repetitive administrative work. Those gains are real, but the more consequential questions are emerging now that AI is moving deeper into recruiting workflows and firms are relying on its outputs rather than merely testing its capabilities.
Some of those risks are becoming harder to audit as research finds that AI models can develop stereotypes and decision patterns that were never intentionally programmed. In one study of dialect prejudice, larger language models exhibited more covert prejudice than smaller ones even as they became better at processing the dialect being studied.
Existing safeguards were largely designed to detect discrimination against categories firms already know to monitor, which poses a challenge as AI produces less familiar forms of bias. New York City’s response has been to require bias audits for certain automated employment tools, although a December 2025 state comptroller review found significant weaknesses in how the city enforced the law. And the European Union is taking a different approach, classifying recruitment AI as high-risk as major AI Act obligations begin taking effect.
Who’s missing?
For search firms, the most difficult blind spot may be hidden in what appears to be a strength: AI can generate a convincing list of candidates very quickly. Recruiters can inspect the people a system recommends and question why they received a particular score. It is much harder to examine the people who never appeared. Who was missed, and why?
AI-generated lists are more useful when recruiters have something independent to compare them against. A recruiter’s own sourcing may turn up people the system missed, just as AI may uncover candidates who would not have surfaced through a traditional search. Comparing the results can help reveal who is missing before an AI-generated list starts defining the candidate pool.
Responsibility also extends beyond the candidate list itself. In an ongoing federal discrimination case, a judge allowed claims against cloud HR software company Workday to proceed under a theory that the software provider could potentially act as an agent of employers when performing traditional hiring functions such as screening applicants. The case has not established that Workday discriminated, but it does show why firms need to think carefully about the vendors whose technology participates in hiring decisions.
That scrutiny has to go below the product surface. Search firms should understand which models power the tools they buy, where candidate information is processed, whether it is retained, who can access it and how recommendations are generated. Those questions carry particular weight in executive search, where longstanding professional standards already impose strict confidentiality obligations around candidate information. Questions that may once have belonged mostly to IT or legal teams now need to be considered before a recruiting platform becomes embedded in day-to-day search work. The more consequential the role AI plays in narrowing a candidate pool, the less defensible it becomes to treat those details as somebody else’s problem.
Even vendor discovery can create another blind spot. Research has found substantial variation in how major AI systems rank products when users submit identical recommendation prompts. A chatbot can help generate names to investigate, but popularity inside an AI-generated list is not evidence that a product is appropriate for a firm’s workflow, risk tolerance or data obligations.
Firms don’t need to respond by pulling back from AI. The technology can broaden sourcing, handle repetitive work and leave recruiters with more time for candidates and clients. But they need to decide where human review is required, preserve enough independent sourcing to catch omissions and scrutinise vendors before their technology becomes part of a process that affects real candidates.
Executive search still doesn’t have a settled formula for getting the most from AI. It does have more evidence of where things can go wrong, from opaque recommendations and missing candidates to poorly understood vendors and responsibility that doesn’t disappear when technology sits between a firm and a decision. Avoiding those failures may be where a useful AI playbook begins.
