In the rapidly evolving landscape of artificial intelligence, a new study has uncovered a troubling bias embedded in popular large language models like ChatGPT: these systems systematically recommend lower salary expectations for women compared to men, even when qualifications are identical.

Researchers at the Technical University of Würzburg-Schweinfurt in Germany tested five leading LLMs, including OpenAI’s ChatGPT, by feeding them identical profiles that differed only in gender. The results, detailed in a preprint paper, showed that female personas were advised to aim for salaries up to 20% lower than their male counterparts in fields like law, medicine, and engineering.

This revelation comes at a time when AI tools are increasingly integrated into professional workflows, from resume drafting to negotiation coaching. According to The Next Web, which first reported on the study, the bias reflects deeper societal inequities baked into the training data of these models. Lead researcher Ivan Yamshchikov noted that a mere two-letter change in prompts—from “he” to “she”—could result in salary advice differing by as much as $120,000 annually, underscoring how subtle inputs amplify systemic discrimination.

Unpacking the Methodology and Broader Implications

The experiment involved prompting the AI with scenarios where applicants sought salary negotiation tips for roles in high-stakes industries. Across thousands of iterations, the models consistently undervalued women’s worth, with the gap most pronounced in male-dominated sectors. TechNews, a Taiwanese publication covering AI developments, highlighted that this isn’t just a quirk of ChatGPT but a pattern observed in models like Google’s Bard and Anthropic’s Claude, suggesting a widespread issue in how LLMs process gender cues.

Industry insiders argue this bias perpetuates the gender pay gap, which persists globally at around 20% according to World Economic Forum data. As companies rely more on AI for HR functions, such flaws could automate inequality. Cybernews reported similar findings, emphasizing that the models’ training on vast internet datasets—rife with historical biases—means they often mirror rather than mitigate human prejudices.

Echoes of Past AI Controversies and Calls for Reform

This isn’t the first time AI has been called out for gender discrimination. A 2019 article from The Next Web discussed a feminist chatbot designed to counter biases in voice assistants, which are often defaulted to female personas. More recently, MIT News explored how ChatGPT boosts productivity in writing tasks but warned of unintended consequences in professional advice.

Experts like Yamshchikov advocate for greater transparency in AI development, including diverse training data and bias audits. WebProNews, in its coverage, stressed the need for regulatory oversight to ensure fairness, especially as LLMs influence career trajectories. OpenAI, the creator of ChatGPT, has acknowledged such issues in past statements, pledging ongoing improvements, but critics say progress is slow.

Toward Equitable AI: Challenges and Opportunities

The study’s implications extend beyond salaries to career advice in general. Macao News noted that LLMs also suggested different goal-setting strategies based on gender, with women steered toward “balanced” life choices over ambitious pursuits. This could subtly discourage female advancement in competitive fields.

For tech leaders, the path forward involves ethical AI frameworks. As Hacker News discussions pointed out, referencing a 2019 MIT Technology Review piece on algorithmic biases, ignoring these problems risks entrenching discrimination. With AI’s role in the workplace growing—evidenced by a New York Times report on workers experimenting with ChatGPT—the urgency to address gender biases has never been higher. Ultimately, building fairer models requires not just technical fixes but a cultural shift in how we curate data for the machines shaping our futures.