When You Can’t Explain a Rejection, You Own the Risk

Erin Kistler has spent four years applying to jobs she is plainly qualified for, and it has gotten her nowhere. According to a Guardian report cited by tovima.com, the product director has applied to thousands of roles at companies including PayPal, Microsoft and Netflix without landing a single interview.

She has nearly 20 years of experience. She is now suing Eightfold AI, the vendor behind the screening software, as part of a class action filed in January in a California court.

It is tempting to read her case as one flawed algorithm’s failure. It is not. It is what happens when a company hands the judgment that decides someone’s career to a system nobody inside the company can actually explain.

The Score Nobody Inside the Building Can Defend

Eightfold describes itself as the largest, self-refreshing source of talent data in the world, continuously updating an internal database from the resumes, LinkedIn profiles and social media of more than a billion candidates who have applied through its platform.

It uses that data to score candidates from 0 to 5, predicting how well they will perform in a given role. That score determines who gets prioritized for an interview and who does not.

Rachel Dempsey, the lawyer representing Kistler, put the problem plainly: “A big part of the problem is that job applicants don’t know what’s in these dossiers.” “The idea of a black box is deeply unsettling,” she said.

Dempsey is right that the candidate cannot see what sits behind that score. In our work with hiring teams, the more common failure sits one step earlier: the recruiter or hiring manager who receives the score usually cannot reconstruct how it was reached, because nobody asked the vendor to show its reasoning before rollout. Ask whether anyone on your team could reproduce that logic if asked. The vendor built the algorithm. Your company owns the decision it produced.

What the Research Behind the Objectivity Claim Actually Found

The pitch for AI screening is objectivity at scale: strip out human favoritism and the process gets fairer. Ifeoma Ajunwa, a professor at Emory University School of Law and founding director of its AI and the Future of Work program, has studied what actually happens instead. “As we’ve done more research on AI hiring systems, we’re finding that they actually tend to reproduce many of the same biases human managers have,” she said.

Xuechunzi Bai, an assistant professor at the University of Chicago, went further. Using fictional demographic groups, her research found AI models stereotyping candidates on traits with no bearing on their qualifications, showing greater bias than the human hiring decisions examined that year. Newer, more advanced models produced even more bias, not less.

That is the finding worth sitting with: tools sold to fix human inconsistency scaled the same instincts across employers that adopted them. According to Ajunwa, “you’ve essentially been placed on an algorithmic blacklist.” Katie Creel, co-author of a separate study on algorithmic exclusion, stressed that people may end up shut out of jobs far more than they otherwise would be.

Three Companies, Three Different Failure Modes

Kistler’s case is not the only one working through US courts. Other workers are suing Meta over an internal AI system they allege targeted them for layoffs after parental or medical leave, and a separate lawsuit against IBM claims its AI tools discriminated against older workers. These are allegations, not findings. All three companies deny wrongdoing, and Eightfold and IBM have called the claims baseless, according to the Guardian’s reporting.

None of these cases has been decided, and a filed complaint is a claim, not a ruling. What the three share, regardless of how a court eventually decides them, is the shape of the exposure: a company let a system make or heavily influence a personnel decision, and when asked to justify it, could not point to a documented, human-reviewable reason.

The Compliance Floor That Skips the Human-in-the-Loop Gap

Ajunwa is blunt about the legal backdrop. “There’s actually no law requiring notice or disclosure of the use of these AI hiring systems,” she said. “So companies don’t always tell employees when AI is being used to evaluate them,” she added.

That matters given how common the practice has become. According to a World Economic Forum report, 90% of employers used some form of automation in hiring last year, ranging from basic filters to AI-run phone screens.

Some vendors already guard against this. Incredible Health keeps AI from making the final call on who advances, leaving a human to approve the decision, an approach several US states now require by law.

New York’s version, which took effect in 2023, requires employers using automated hiring systems to run annual bias audits and notify candidates in advance, but only for software that substantially assists or replaces decision-making. That leaves a gap for employers who keep a human formally in the loop while letting that human defer entirely to the system’s score.

A law built around whether a human technically remains in the process says nothing about whether that human can account for what the process actually did.

The lawsuits against Eightfold, Meta and IBM will take years to work through, and none of the allegations has been proven. The risk does not wait on a verdict: in our view, any employer using a vendor score to decide who gets an interview is already answerable for the reasoning behind that score, whether or not anyone inside the company has read it.

Original reporting: tovima.com.