An artificial intelligence (AI) tool reported inaccurate grades for medical students applying for residency positions, sparking concerns among the students and AI scholars.

In late September, Thalamus Cortex, an AI technology that assists in residency application review and screening, incorrectly transferred some grades from students’ transcripts to a summary page that highlights applicants’ key information and characteristics.

Thalamus CEO Jason Reminick, MD, MBA, MS, told Medscape Medical News that the number of students with affected grades was low and that fewer than 0.7% of grades extracted were inaccurate. In a blog post, he said Thalamus had received “just 10 reported inaccuracies out of over 4000 customer inquiries this season.”

Medical AI experts said the incident highlights the perils of using AI to make high-stakes decisions.

“This is the risk of these sorts of systems. In AI, all language models hallucinate. There’s no way around it,” said Adam Rodman, an internist and director of AI Programs for the Carl J. Shapiro Institute for Education and Research at Beth Israel Deaconess Medical Center in Boston.

In his blog post, Reminick said the incorrect grades came about not because of “AI hallucination” but by optical character reading variances. He noted that about 10% of residency programs used Thalamus in evaluating candidates this year.

Panicked Students

Several fourth-year medical students told Medscape Medical News they had panicked over the possibility that the AI tool had reported lower grades than they received.

“It’s already a tough enough process for medical students. And now in the back of your mind, you’re thinking, like, ‘Did I not get that one interview because they got a wrong transcript?’ Yeah, it’s a small percentage, but if you have no way of knowing whether you’re impacted or not, it just creates a lot of anxiety for everyone,” said one fourth-year medical student in New Jersey. MedscapeMedical News is not using this student’s name because he fears retaliation.

This fourth-year student, who applied to more than 50 residency programs, said he received no direct communication from Thalamus about the issue. Thalamus on October 6 published an initial blog post on the incident, which is when some medical students said they first learned about it.

The New Jersey student said his medical school and a residency program director alerted him to the grade discrepancy. The program director told the student they had noticed a disconnect between the grades on the student’s downloaded transcript and the ones listed in Thalamus and had informed the student’s medical school.

Reminick said the key data taken from residency applications and transferred to Thalamus are more than 99% accurate for this season and insisted that students’ residency program decisions should not be affected by the malfunction.

“This was a very, very contained number of students,” said Reminick. “We have not been able to find evidence that any faculty member has ever taken a single inaccurate grade as the deciding factor in any students across the platform to date.”

Risk for Inaccuracies

Medical doctors who are also experts in AI told Medscape Medical News that using AI to evaluate residency applications can be dicey if it’s not held to a high standard of accuracy and transparency.

“In something as high stakes as residency selection, I don’t think that this type of technology, which has a nonzero error rate, should be used,” said Rodman.

If an AI software incorrectly extracts information that’s never corrected, a student could potentially miss out on a residency interview.

“I think that any use of an AI tool is potentially problematic if it’s not thoroughly vetted,” said Shivam Vedak, MD, MBA, a clinical assistant professor in the Division of Hospital Medicine at Stanford University School of Medicine, who researches generative AI in clinical settings.

Vedak said based on his analysis of Thalamus’ Cortex model and the company’s blog posts explaining their software and methodology, it appeared to him “relatively well evaluated and embedded,” but he cautioned the technology should be held to an extremely high standard.

“I think that one of the missteps was actually about transparency,” said Vedak. “In this case, especially with something as experimental and in a way controversial as AI, Thalamus should be disclosing to set the medical students at ease in how well this is vetted, and so students don’t end up filling that void with things they might have heard that may be misinformation.”

AI Already Baked Into Residency Applications

The Association for American Medical Schools (AAMC), which represents US medical schools and manages the residency application system, first started a partnership with Thalamus in 2023 to streamline the residency interview system.

Thalamus’ initial software was a scheduling platform that helps students and programs to organize residency interviews in one integrated system. Reminick likened the software to booking a restaurant reservation but instead for interview scheduling for applicants.

“At the time, there were multiple tools being used by programs to interview applicants, and applicants were juggling all their interviews across multiple platforms,” said Patrick Fritz, senior director for residency and fellowship services at the AAMC.

This 2025 application cycle is the first time the Cortex software has been offered for free to residency programs, through the Thalamus partnership with the AAMC.

The AAMC and Thalamus told Medscape Medical News public webinars were held so medical students and residency programs could better understand the Cortex software, but several students said they weren’t initially aware that residency programs were using it.

“In hindsight, it’s clear many students want more visibility into how their data is displayed. We apologize to those who believe that more widespread communication would have been helpful,” Reminick wrote in an October 16 blog post on the Thalamus website.

Reminick has pledged to fast-track a medical student portal that will allow students to view the Cortex page that contains the AI summary of their residency application, which should be available for next year’s residency application cycle.

More Accountability for AI

“These program directors, these deans of admissions for medical schools, are essentially tasked with screening hundreds, if not thousands of applications a year, right? And these applications are not trivial. It’s a little bit of a Sisyphean task,” said Dong-han Yao, a physician informaticist and emergency physician at Stanford University. “I certainly don’t think we are going to go backwards from here.”

But medical schools and residency programs need to demand more accountability from the companies that make the AI tools they use, Yao said.

“The responsibility of monitoring (AI tools) to make sure that there is no harm is sort of incumbent on both the company themselves, but also, we as medical schools or the people on the receiving end, need to hold them accountable in some way as well,” said Yao. “The task itself of evaluating applications is mission critical, and so it needs to be held to an extremely high standard.”

Victoria Knight is a freelance reporter in Washington, DC.