In a world transitioning to a reliance on artificial intelligence (AI), the future of the job market is following suit. As of now, it seems that there are more harms than benefits of utilizing AI for talent acquisition and recruitment.  

Biases in AI-Powered Hiring 

An estimated 99% of Fortune 500 companies use some form of automation to screen or rank candidates for hire. Nearly one in four medium-sized employers uses AI in their hiring process, according to recent surveys from the Society for Human Resource Management. The hiring process that was once a time-consuming and human-driven process of recruiters going through resumes, conducting interviews, and making decisions based on experience has shifted to utilizing AI to supposedly boost efficiency, maximize reach, and decrease human bias in the recruitment process. 

Many studies show that this claim is a misconception, with AI and automation causing more bias and minimizing outreach. Researchers at the University of Washington found that large language models (LLMs), which are large deep learning models that are pre-trained, favored white-associated names 85% on over 550 real-world resumes that included names with white and Black men and women. Black male-associated names were never favored over white male-associated names. 

Additionally, a Bloomberg analysis found that OpenAI’s ChatGPT “systemically produces biases that disadvantage groups based on their names” when they ranked resumes with different names that are demographically distinct. 

However, a study from the London School of Economics and Political Science contradicts this claim, stating that AI hiring resulted in more diverse outcomes than human hiring and was more “fair.” The study also noted candidates' and recruiters’ overwhelmingly negative reaction to AI hiring. 

Furthermore, algorithmic models use keywords of a model applicant to identify the most talented individuals who align with the company’s values. Although the automation is time-efficient in theory, the keyword-based matching system puts those with unconventional job titles and a non-linear path at a disadvantage. This system reduces the diversity of thought and experiences. Algorithmic “black boxes,” where internal systems are not easily understandable, offer little insight into the decision-making process, making the biases rooted in algorithms difficult to identify and change. 

The black-boxed algorithm could lead to “technological redlining,” where systemic exclusions are perpetrated by inequitable outcomes and replicate known inequalities from algorithms, according to UCLA professor Safiya Noble. Incorrect or incomplete data inputted by humans can create biased output, reflecting unintended cognitive biases or real-life prejudices. This bias could create an unethical cycle where AI machine learning algorithms train and validate their systems based on faulty data. 

The racism rooted in the algorithms is, once again, difficult to attribute and solve. In addition to algorithmic inequality, another type of digital redlining could be produced by identifying and discriminating against certain locations, device models, and potentially internet behavior. 

Therefore, it is proven that current models of AI cannot eliminate human bias through automation; in fact, AI algorithms may create more unwanted discrimination based on biased human input data and algorithmic faults. With a lack of human oversight, it could be difficult to minimize the harms of biased AI. Even trying to reduce bias according to one definition could invariably result in increased bias according to another definition, which was shown through the Amsterdam Smart Check project, which attempted, but failed, to create an ethical AI system. 

Potential Legal Concerns and Costs 

There is also the potential liability of legal concerns. Even unintentionally, discriminatory practices through the usage of automated AI have resulted in lawsuits. For example, in the Mobley v. Workday District Court case, which occurred in early July 2024, the plaintiff claimed that Workday’s AI-powered applicant screening tools discriminated against him based on race, age, and disability. While the ruling dismissed the plaintiff’s claim, the case showed that AI vendors and AI-powered hiring tools could be held liable under anti-discrimination laws.  

As Mobley v. Workday was the first major U.S. case to permit direct liability for an AI service provider, there is little data as to how much companies could be charged in an AI discrimination lawsuit for using AI-powered hiring tools. However, depending on the scale of the suit, lawsuits with many applicants, legal fees, and compensatory damage payments could lead to hundreds of thousands, if not millions of dollars, in reparations. 

Although AI-powered hiring has drawn skepticism amongst its users, it is important to note the efficiency and potential benefits of the model. According to a study from the London School of Economics and Political Science, AI hiring “improves efficiency in hiring by being faster, increasing the fill-rate for open positions, and recommending candidates with a greater likelihood of being hired after an interview.” 

In addition to recruiters using AI-powered tools for the hiring process, a large number of job candidates are using AI to apply to jobs; in fact, 65% of job candidates are using AI at some point in the application process, according to the 2025 Market Trend Report from Career Group Companies. This includes writing cover letters, drafting resumes, headshot alterations, interview practice, and career guidance. 

Since eliminating AI bias seems unrealistic with the current skillsets of algorithms, the first step to mitigate discrimination in code should begin with the individuals who design and train the AI systems, especially if the models are being used to address and rank human profiles in the job industry. Perhaps companies that decide to utilize AI in their hiring process should proactively discern where the AI models are coming from and who created them in order to reduce future discrimination and liabilities. Although there are benefits to thriving in an anti-monopoly market with accessible open source code, there is value in regulating the data that is spoon-fed to AI machine learning models in order to reduce prejudice and bias.