Are coroners and medical examiners already using artificial intelligence (AI) in their medicolegal death investigations? For what purposes? How will AI affect the work of coroners? It’s time to ask these questions.
In 2022, a scientific paper claimed that “the traditional way of doing an autopsy and framing an opinion has a lot of limitations and these limitations can be overcome with AI.”
Where are we with that three years later?
In 2025, Johns Hopkins University (JHU) held a symposium on The Future of Forensics: How AI Can Transform Investigations.” The overall tone of the symposium reflected an abundance of skepticism and caution about the legal aspects of AI and forensics.
I was puzzled to find no mention of coroners at the JHU symposium, so I dug deeper, looking at four potential targets for AI use in death investigation: autopsies, toxicological testing, cause of death determination, and manner of death determination.
AI and autopsies
The ongoing shortage of forensic pathologists (FPs) offers opportunities and incentives for AI deployment.
Autopsy documentation
An academia-commercial partnership recently presented early results of a transcription program customized for use during autopsies. The developers describe “AutopsyAI” as “an application that enables real-time, structured autopsy documentation with embedded artificial intelligence (AI) support for generating and organizing autopsy reports.”
Postmortem CT scan interpretation
Because of the shortage and expense of FPs, coroners sometimes use postmortem CT scans in place of an autopsy. Usually, it’s still a pathologist who reads those scans, but pathologists are not radiologists. Errors can occur.
This might be a potential application for deep learning AI models. However, the field is in its infancy. A 2025 study looking at body composition in cadaver CT scans focused on only two endpoints: age determination, and whether the cause of death was acute or chronic. This kind of information comes nowhere near the specificity needed for a cause of death.
Still, image analysis (death scene & autopsy photos, CT scans) may be the low-hanging fruit for AI in forensic pathology. Analysis of gunshot wounds is an area of ongoing research.
AI in analysis of gunshot wounds
A 2021 proof-of-concept study assessed the ability of AI to determine shooting distance. The researchers used known distances (piglet carcasses, not humans!) to train a deep learning model. The model achieved an impressive 98% accuracy based on analysis of over 200 photographs of the wounds. Then again, even though I’m not a forensic pathologist, I had little trouble differentiating between the photographs of contact, close-range, and distant shots.
A 2025 review article in Frontiers in Medicine included eighteen research papers (1990-2025) on use of AI in forensic pathology. Possible uses included pattern recognition. Injury analysis, including gunshot wounds, showed high accuracy. Specifically, one study used a deep learning system that “correctly classified 88.19% of entrance wounds and 87.71% of exit wounds.” That study used the assessments of human forensic pathologists for comparison. Humans are never perfect, but neither is AI.
The authors of the review concluded that “artificial Intelligence serves best as an enhancement rather than a replacement for human expertise.”
In October 2025, researchers presented two abstracts on the topic of AI analysis of gunshot wounds. The consensus seems to be that machine learning AI tools show promise in injury assessment, but are far from practical implementation.
AI and postmortem toxicology
The bottleneck in determining the cause and sometimes the manner of death is usually toxicological testing. If results could be available sooner after an autopsy, the average time to final certification would be shorter.
There are ongoing efforts to use AI tools for predicting toxicology of, for example, chemical compounds (think cosmetics) or pharmaceuticals. But I couldn’t find any information about its application in speeding up postmortem specimen testing.
I did find one study that used a natural language processing–based model in an attempt to predict “the likelihood that an autopsy report narrative describes an accidental or undetermined fatal drug overdose.” Obviously, it’s one thing to “predict” a drug overdose. It’s quite another to produce evidence that will hold up in court.
Bottom line: AI is not going to speed up death certification anytime soon.
AI and cause of death determination
Determining cause of death is an essential coroner function. Efforts are underway to insert AI into this core activity.
According to a 2024 report, the government of Greece was on the case with a draft bill that would provide coroners with “a ‘digital assistant'” that will help the medical evaluation by providing probable causes of death based on the findings and a consultation with the relevant precedents and bibliography done at a speed unmatched by humans.”
Just last month (November 2025), after an investigation found “long-standing failures in autopsy procedures,” the 2024 proposal became a reality. Greece will have not just “digital assistants,” but new morgue facilities.
In China, researchers have developed a large language model (LLM) designed to analyze autopsy findings and develop “concise cause of death” statements. The developers took pains to include “human-in-the-loop feedback to ensure legal and medical validity.” You can read more about it in this LinkedIn post.
See also the section on AI and autopsies above.
AI and manner of death determination
Arguably, determining the manner of death is the most sensitive and the most subjective task of a coroner. The lives of survivors and the fates of those accused of homicide hang in the balance.
My limited online search provided no information about using AI for manner of death determination. Maybe that’s a good thing. Maybe it’s just a matter of time. But everything I read emphasized the continued need for human involvement.
Legal Validity of AI
Coroners and medical examiners are often called to testify in court. Several years ago, I was a witness in a drug death trial. The defense attorney asked me a question during cross-examination: “It is YOU, isn’t it, who determines the manner of death?” At the time, I considered it a rhetorical question. My answer—yes—came with no hesitation.
The attorney didn’t ask me how I’d come to my conclusion about the manner of death. That, it seems to me, would have been the more interesting question.
In the not-too-distant future, coroners may be using AI bots to help determine the manner of death. Perhaps they will input what’s known about the case, then get a response from a generative AI program that lists the probabilities of the various manners of death. I wouldn’t be surprised if some coroners are already doing that.
When such a future comes to pass, will the coroner have to explain the AI technology and its sources to defense attorneys and juries? In order for DNA or toxicological test results to stand up in court, coroners must use certified laboratories. Standards for certification of AI bots must be developed before their output can be legally admissible in court. Right now they don’t exist.
“AI has promise in medicolegal death investigation, but it’s got a long way to go.“
Future directions for AI in death investigation
Just two months ago, a New Zealand group published a paper describing a framework for forensic-science based AI projects. This is an essential first step in ensuring projects are carried out within a scientifically valid and ethical framework.
Here’s a short list of projects in-progress and a few at the wishful thinking stage.
Data extraction and analysis
Data is arguably the most important “product” of a coroner office. Most of that data is hidden within disparate case management systems and/or in the “natural language” used in case narratives, autopsy reports, and death certificates.
As coroner, especially during the COVID-19 pandemic, I did my best to manually extract information so I could share insights with the public and various stakeholders. I might have welcomed a tool like MDI Connect. According to an abstract presented at the 2025 National Association of Medical Examiners Annual Meeting, MDI Connect is a collaborative CDC-led project. Its goal is that “data (in coroner and medical examiner offices) will become easily searchable and more accessible for analysis and sharing.”
AI-assisted CT scans
CT scans generate very detailed images, so an AI model trained and calibrated on postmortem images might be a way to bridge the gap. For coroner offices in more remote rural counties, AI-assisted scans might be a cost-effective solution to limited access to forensic pathologists.
AI and environmental or disaster deaths
I’ve pointed out the variability of heat death reporting in different locations in the United States, so I’d especially love to see AI analyzing such cases using a standardized assessment tool, then comparing the determinations of different forensic pathologists.
The same study design could be used for other controversial deaths, such as deaths in custody.
AI and decedent identification
Identification through facial recognition technology will, like DNA, depend on whether the decedent’s face is already in a database. But it’s likely all our faces will be in one database or another in the near future. That means that AI facial recognition technology, like fingerprints, may be able to identify someone right at the death scene.
Conclusions about AI in death investigation
The use of AI in medicolegal death investigation is in its infancy. It is not currently in practical use at coroner or medical examiner offices, at least not in the United States.
The big questions revolve around
- training (what data is available, and what about privacy concerns?)
- proving reliability across the broad spectrum of medicolegal death investigations
- legal admissibility of AI-assisted forensic determinations
- cost (not discussed in this blog or in papers I read, but probably the first issue that will be raised by every local government)
Take-home message: AI has promise in the world of medicolegal death investigation, but it’s got a long way to go. In the short term (I don’t know what that means these days…), real humans will still perform the autopsies, sign the death certificates, speak with grieving next of kin, and testify in courts.