Can AI help police investigators solve cold cases?
Despite the best efforts of tireless police detectives and law enforcement investigators nationwide, the United States is sitting on an estimated 250,000 to 300,000 unsolved homicides.
Every year, another 6,000 to 7,000 murders go unsolved, quietly joining the backlog. In the 1960s, police cleared roughly 90% of homicides. Today, the national clearance rate hovers around 50–55%, and in some major cities (like Chicago) it is far lower.
That means every year, thousands of families move from shock to silence without answers. Without hope. And homicide is only part of the story.
At any given moment, 80,000 to 90,000 people are listed as missing in the United States. Many of those cases resolve quickly. But a smaller and more haunting category — estimated somewhere between 15,000 and 25,000 people — becomes long-term, unresolved disappearances.
Then there are sexual assault and violent crime backlogs. For decades, hundreds of thousands of rape kits went untested. Many departments still carry unsolved assault, kidnapping, and violent felony cases that date back years — or decades. There is no precise national aggregate for all of these combined, but when you begin adding them together, the number of serious cold cases likely rises into the 300,000 to 400,000 range, at minimum.
Some criminologists call that estimate conservative.
Detectives sometimes label cases as having no viable suspect, no usable forensic evidence, or no available witnesses. But “unsolvable” cases are rare. Frequently, cold cases happened because of the limited resources and technology available to investigators at the time.
But times have changed.
DNA has reopened cases once considered hopeless. Forensic genealogy has identified killers decades later. New fingerprint techniques, digital analysis tools, and renewed witness cooperation have resurrected files that once gathered dust.
Criminologists who study cold cases often estimate that 10 to 30 percent of unsolved homicides could potentially be solved if reinvestigated with modern tools. Even if we take the cautious end of that range — say 15% of 250,000 unsolved murders — that would mean roughly 37,000 potentially solvable homicide cases sitting in archives across the country.
Thirty-seven thousand families who might yet receive answers.
And that number does not include sexual assault cases reopened through DNA, missing persons cases aided by new databases, or violent felonies linked through modern data analysis.
The hard reality is that many cold cases are not mysteries in the dramatic sense. They are under-resourced. They are buried under active caseloads. They are casualties of turnover, budget constraints, and the simple fact that human beings cannot re-read ten thousand pages of evidence every time a new lead appears.
This is not a failure of dedication. It is a failure of scale.
And that is precisely where a tool like artificial intelligence becomes relevant.
The opportunity is not magic. It is more like triage.
If even 10 to 20 percent of cold cases are structurally solvable — meaning evidence exists but has not been reprocessed, cross-referenced, or viewed through modern analytical tools — then the ability to organize massive datasets, detect hidden patterns, and resurface overlooked connections could matter enormously.
Every cold case has two enemies: time and scale.
Time corrodes memory. Witnesses move. Detectives retire. Evidence degrades. The emotional urgency that once drove a newsroom or a precinct fades into archive boxes and mislabeled hard drives.
But scale is the quieter enemy. Modern investigations generate oceans of data — phone records, surveillance footage, license plate readers, DNA reports, financial transactions, tip lines, interview transcripts. When a case goes cold, it’s rarely because there is nothing there. It’s because there is too much.
That’s where artificial intelligence enters the story — not as a robot detective, but as a sorting machine.
Imagine a homicide file from 1997. Ten thousand pages. Hundreds of tips. Handwritten notes scanned into PDFs. Interviews conducted by three different detectives across two counties. Somewhere inside that pile is a witness who mentioned a red pickup truck with a cracked tail light. In another file from 2001, a similar vehicle appears in a seemingly unrelated assault. No one ever connected them because no one had the bandwidth to read everything side by side.
The most powerful applications for law enforcement are almost boring. AI can summarize decades-old case files so a new detective doesn’t spend six months just “getting up to speed.” It can re-rank thousands of tips and identify the ones that share geographic or behavioral overlap. It can help reconstruct timelines from digital breadcrumbs — phone pings, financial transactions, GPS logs.
It can also enhance degraded audio and video, helping investigators extract usable information from recordings that were previously too noisy or too grainy to analyze effectively.
In that sense, AI is less “RoboCop” and more “archivist on steroids.”
Cold cases deserve answers. Victims’ families deserve closure.
AI may become one of the most powerful investigative tools ever created.
The real test will be whether we are disciplined enough to use it without letting it use us.
(Contributing writer, Brooke Bell)