Can the Internet Still Solve Crimes When Evidence Can Be Fabricated?

Content note: This article discusses murder and false accusations.

The internet turned ordinary people into amateur investigators who could spot clues professionals missed. Now those internet sleuths are working in an era when the evidence itself can be manufactured.

By Jack Malacat

For years, internet sleuths have been one of the stranger byproducts of being permanently online. A stranger can freeze a frame and match a sliver of background to a real street, or catch a detail that thousands of other people scrolled past. The Netflix documentary Don't F**k with Cats made that kind of collective hunt feel cinematic, following a group of online volunteers who spent months trying to unmask the man behind a series of animal-cruelty videos, later identified as Luka Magnotta. What once required police files or specialist tools can now begin with a laptop and someone stubborn enough to keep looking.

Sometimes that attention produces something investigators can use. In September 2021, as the search for Gabby Petito drew national attention, the travel vloggers Jenn and Kyle Bethune went back through GoPro footage from their trip through Wyoming. Weeks earlier, a camera they had forgotten to switch off had recorded a white van with Florida plates parked in the Spread Creek camping area, and it looked like hers. They sent the clip to the FBI. Her remains were found in that area on September 19, the same day the video went public. Investigators were already searching Spread Creek, and the FBI never said how much the footage mattered, though its closing update called the public's help invaluable. The clip had sat quietly on a family's hard drive until someone understood what it showed. Ordinary people are constantly making accidental records of moments that only matter later, and that may be the internet's most useful investigative asset.

The same attention can turn a bad guess into something that feels like evidence. After the 2013 Boston Marathon bombing, Reddit users combed through crowd photos looking for the bombers. On the night of April 18, the name of Sunil Tripathi began circulating as a suspect. He was a Brown University student who had been missing for about a month, and he had nothing to do with the attack. One tweet from the hacker collective Anonymous pushed his name to more than a million followers at 3 a.m. The rumor collapsed later that morning, once the real suspects were named. For those few hours, repetition did the work that evidence should have done, and a wrong answer moved with all the momentum of a right one.

Fabricated evidence is much older than artificial intelligence. In November 2016, Sherri Papini disappeared in Northern California and turned up 22 days later on a roadside, injured, saying two Hispanic women had abducted her at gunpoint. The story sent investigators looking for two women who never existed. Federal agents eventually found that she had been staying with a former boyfriend in Southern California and had staged her injuries herself. She was sentenced to 18 months in 2022. Her hoax needed no technology at all, only enough physical traces to make other people take it seriously.

What has changed is how many of those traces can now be made on a screen. In January 2024, an audio clip spread online that seemed to capture the principal of Pikesville High School in Maryland making racist and antisemitic remarks. Angry calls flooded the school, the principal was removed from his post during the investigation, and threats followed. Police later said the voice was generated with AI by the school's athletic director, and experts who examined the file found synthetic speech with background noise layered in by hand. In April 2025, the athletic director took a plea deal and was sentenced to four months. By then the recording had already done its damage. A fake doesn't have to survive scrutiny to hurt someone. It only needs a head start.

Professional investigators have more to work with than someone scrolling a feed. A suspicious file can be checked against the device that supposedly made it and against other records from the same moment. Courts add a gate of their own: under Federal Rule of Evidence 901, a recording can be admitted only when someone offers enough support to show it is what it claims to be. Technologists are building provenance standards, such as the Content Credentials developed by the C2PA, which can travel with a file and record how it was made or edited. Each of these helps, and none of them settles the question alone. Metadata is often stripped when files are shared, and most cameras and platforms don't carry provenance yet. NIST researchers who evaluate deepfake detectors have warned that tools scoring above 90 percent in academic tests often stumble on real forensic casework. Every new generation of fakes reopens the contest.

A stranger problem appears when the mere possibility of fakery becomes an argument. In 2019, the law professors Bobby Chesney and Danielle Citron described what they called the "liar's dividend": as people learn how convincing synthetic media can be, anyone caught on a real recording gains an easy way to call it fake. Courtrooms have already heard the move. In a 2023 wrongful-death lawsuit over a fatal Tesla crash, the company's lawyers suggested that a 2016 video of Elon Musk discussing the cars' self-driving abilities might be a deepfake, since he is a frequent target of them. The judge called the position deeply troubling and ordered Musk to answer questions under oath. Two defendants tried a version of the same argument about footage from the January 6 Capitol riot. Online, there is no judge. A real video can be waved off with a single word, and the doubt spreads on its own.

The internet will probably keep helping solve crimes, because its oldest advantage hasn't gone anywhere. There will always be forgotten footage, and someone willing to spend a night studying what everyone else ignored. What has shifted is the order of the work. A clip that once answered a question may now raise a new one first, about where it came from and whether the moment inside it ever happened. Finding the evidence may turn out to be the easy part.

Will solving crimes eventually require proving that the evidence is real before we can even ask what it proves?

About the author

Jack Malacat is the creator of Venom Haus and the editor of The Venom Haus Archives, a one-man publication made possible with the help of AI. When he isn't designing games or dreaming up new products, he's here, chasing questions about the darker side of being human. He thinks the internet has enough AI slop, and he wants to show what these tools can do in the right hands, with the sources to back it up.

Sources and Further Reading

Lewis, Mark (director). (2019). Don't F**k with Cats: Hunting an Internet Killer. Netflix. The documentary series that followed online volunteers trying to identify the man behind a series of animal-cruelty videos, and helped popularize the modern image of the internet sleuth.

Fox 13 Tampa Bay. (September 2021). "Gabby Petito's possible remains found near where Tampa family spotted camper van." Jenn Bethune's account of finding the van in footage her family recorded by accident.

Federal Bureau of Investigation, Denver Division. (January 21, 2022). "FBI Denver Provides Final Investigative Update on Gabrielle Petito Case." The FBI's closing update, including the date the remains were found and its thanks to the public for thousands of tips.

Davis, Edward F., III; Alves, Alejandro A.; and Sklansky, David Alan. (March 2014). Social Media and Police Leadership: Lessons From Boston. New Perspectives in Policing. U.S. Department of Justice, National Institute of Justice. A review of social media during the Boston Marathon bombing investigation, including how the false identification of Sunil Tripathi spread overnight.

U.S. Attorney's Office, Eastern District of California. (September 19, 2022). "Sherri Papini Sentenced to 18 Months in Prison for Lying to Federal Agents About Being Kidnapped and Defrauding the California Victim Compensation Board." U.S. Department of Justice. Federal documentation of the staged kidnapping and the investigation that exposed it.

Associated Press. (April 2024). "Athletic Director Used AI to Frame Principal With Racist Remarks in Fake Audio Clip, Police Say." Reporting on the Pikesville High School case, including the police account of how the fake audio was made.

Baltimore Banner. (April 2025). Coverage of the plea and the four-month sentence that closed the state case.

Federal Rules of Evidence, Rule 901. "Authenticating or Identifying Evidence." The federal standard requiring evidence sufficient to support a finding that an item is what its proponent claims it is.

Coalition for Content Provenance and Authenticity (C2PA). (2025–2026). C2PA Technical Specification and Content Credentials. The open standard for attaching verifiable information about where a piece of media came from and how it has been changed.

Guan, Haiying; Horan, James; and Zhang, Andrew. (2025). Guardians of Forensic Evidence: Evaluating Analytic Systems Against AI-Generated Deepfakes. Forensics@NIST 2024. National Institute of Standards and Technology. NIST research on how well deepfake detection systems hold up outside the conditions they were trained on.

National Institute of Standards and Technology. (ongoing). Guardians of Forensic Evidence program. NIST's program to build realistic benchmarks and guidance for forensic examiners testing deepfake detectors.

Chesney, Bobby, and Citron, Danielle. (2019). "Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security." California Law Review, 107. The paper that described the "liar's dividend."

Bond, Shannon. (May 8, 2023). "People are trying to claim real videos are deepfakes. The courts are not amused." NPR. Reporting on Tesla's deepfake argument about a real Elon Musk video and similar claims by January 6 defendants.

Madrigal, Alexis C. (April 19, 2013). "It Wasn't Sunil Tripathi: The Anatomy of a Misinformation Disaster." The Atlantic. Further reading: a same-day reconstruction of how the false identification spread.