Human resources departments are becoming less about humans and more about artificial intelligence, with companies opting for software to complete an increasing number of tasks, including hiring employees. But the trend of injecting AI everywhere has some wondering whether the software does more harm than good in candidate selection.
AI has become more involved in the job-hiring process, with people admitting to using tools to write their resumes and apply to hundreds of jobs at once. It’s shifting to employers who are using models to record, summarize and score applicants, something ethicists told Straight Arrow is a growing problem.
Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™
Point phone camera here
Reports from AI software companies show that more U.S. employers are integrating these platforms in their hiring processes at higher rates than before. Dan Myers, a computer science professor at Rollins College in Florida, told Straight Arrow that it’s not uncommon for AI to be used for candidate screening, but allowing the models to make hiring decisions creates an ethical and legal dilemma for employers.
“How can you be really certain that your machine learning model is not accidentally training itself to pick up on some signal that you really should not be considering in the hiring process?” Myers said.
One system dealing with that question is Workday. A lawsuit accuses the company of denying jobs to people over 40, a claim plaintiffs say amounts to age discrimination. The civil lawsuit is in the discovery phase and could affect thousands of U.S. companies who use the software and millions who applied.
“If you’re using a certain model,” Myers said, “you have an obligation to make sure that the way that the model processes its data, the types of conclusions that it draws are defensible and ethical ones, even if not strictly as a matter of law, but as a matter of professionalism and ethics.”
Candidates may know their data is being used for the interview, he added, but may have less transparency on what will happen to the data once the interview is completed. He said an ideal situation would be not to use a person’s likeness to train a model, but employers’ contracts could say otherwise.
The rush to add AI in the hiring process is an apparent response to candidates who have used models like ChatGPT, Claude and others to write their resumes or cover letters, Myers said. Additionally, employers are drowning in applications from AI-assisted applicants spamming hundreds of job openings and need assistance filtering out candidates.
“We’re seeing the intersection of the abilities of AI combined with the fears that people have of being left behind,” he said.
Models can output wrong, incorrect information
All who spoke to Straight Arrow acknowledged a public hesitancy toward AI adoption, while also not wanting to be left behind in technological advances.
“Whether we solve all these problems or not, it’s clear that AI is solving enough problems quickly enough that it is going to be adopted,” James Brusseau, a Pace University philosophy professor and AI ethicist, said.
Two of the surveys highlighted came from TestGorilla and Fellow. TestGorilla ironically bills itself as software that can weed out AI-generated resumes and interview candidates on its AI platform. Fellow says it’s a secure AI meeting recording, transcribing and summarization platform.
TestGorilla said in a 2025 report that 70% of U.S. employers have used AI in the hiring process. Fellow surveyed an undisclosed number of professionals in 2025 and found that about 75% used an AI notetaker during meetings. Of those, about half noticed the software recording or sharing something that wasn’t intended to be captured.
Changing how a software carries out instructions isn’t an easy feat since AI, like humans, can make mistakes, Brusseau said. Instead, people should focus on using models within their limitations. If a model cannot give an exact summary of a meeting, but can get transcriptions close to perfect, then it shouldn’t be relied on for perfect summaries.
The same line of thinking applies in how software selects candidates, he said.
“It’s impossible to know beforehand exactly why an AI will decide that one candidate is better than another; however, that said, the tools do seem to have some degree of success in identifying candidates who will be successful,” he said.
Opting out of recordings may not have much effect beyond personal choice. Brusseau said the models are being fed so many data points to work in the first place that objecting has a minute effect.
“Any one person’s contribution to the final functioning AI tool is really negligible,” Brusseau said.
Rhongho Jang, an assistant professor of computer science at Wayne State University, told Straight Arrow that among the ethical concerns people have over their likeness being used to train models is whether companies run software locally. If that’s done, he said a candidate’s data never leaves the company.
But smaller or mid-sized companies may opt for cheaper software that instead stores information on the software company’s servers.
“The current technological progress to protect privacy is still on the reliable side, rather than raising great concerns,” Jang said.
AI still has a valuable purpose
The professors said AI tools can be valuable in the hiring process. It helps managers screen candidates and allows them to tackle the more complex parts of interviewing. But guardrails and an understanding of the models’ limitations should be essential.
Brusseau brought up a study where an AI tool was tasked with correctly identifying an Alaskan husky apart from other dogs. The model performed well, but it wasn’t identifying the dogs. Instead, it deduced that if a photo had snow, the dog pictured was a husky.
“There’s no way to know beforehand, or actually even afterwards, in most cases,” he said, “exactly what it is they are picking up on, or why it is they are making the decision one way or another about a certain candidate.”
It could be possible to reverse engineer software that won’t draw unsuitable or inaccurate interview summaries, Jang said. He said there’s a field called machine unlearning focused on removing bad or inefficient data from models, but scientists aren’t entirely certain whether data could be completely removed.
Jang is more concerned about the overuse of AI tools, eliminating the personable aspect of job interviews where candidates and employers learn about each other’s characteristics beyond job performance.
The applicant field’s response depresses Jang. Having learned what the AI screening programs look for, they’ve begun to sanitize their resumes and interview answers to be accepted by the program. It establishes a sense that people should be demotivated to show off their personalities, which Jang said harms candidates and recruiters.
“The AI-assisted interviewing process cannot fully catch the personalizations or characteristics that human beings know is the most important variable a human being can have,” he said.
Round out your reading
- Lawyer says 57 Cornell frat members were in the group chat. Could their silence be a crime?
- Florida county discovers mysterious Flock cameras with no obvious owner.
- Off Script: They won their primaries on one message. Now it’s disappearing.
- Some question whether a popular affordable housing policy is making the nation’s home shortage worse.
- Kelvin wave incoming: Why the US should care.
