The Two Faces of Invisible AI: Catching Criminals, and Wrongly Targeting People

Artificial intelligence is being used on both sides of the criminal battlefield. Criminal groups deploy voice alteration, deepfakes and hacking tools, while investigators and government agencies use AI to analyze video, identify suspicious transactions, defend against cyberattacks and detect manipulated media.

A case in Seoul illustrates the promise of these systems. In January 2026, a man followed a woman in a Gangdong district alley, knocked her down and caused a fracture. He denied intending to commit the act, but prosecutors used forensic analysis, including an AI video-analysis model, to conclude that he had deliberately pushed her. They indicted him in June.

The use of AI in crime prevention has been discussed since the 1990s and became more concrete after machine-learning concepts and deployable software developed around 2016. Financial regulators, forensic investigators and cybersecurity companies began introducing systems that learn recurring patterns in criminal activity and flag suspicious behavior. AI is also being used against voice phishing, as criminals use synthetic voices to deceive victims and authorities analyze calls and victim behavior to limit losses and prevent attacks.

“

Even if ten criminals escape, one innocent person must not suffer.

But the same systems can create new forms of injustice. The article recalls the principle that even if ten criminals escape, one innocent person must not suffer. Current AI cannot fully remove bias from its data, designers or users, yet people increasingly treat its judgments as nearly infallible.

In the United States, a woman in her 50s in Tennessee spent five months in jail after investigators identified her as a leading suspect in a financial fraud case in an area she had never visited. A man in his 20s in Georgia was detained for about a week because his social-media photographs resembled those of a suspect. In both cases, the investigation relied on AI facial-recognition analysis.

AI has also complicated plagiarism checks and music production. Critics say some screening tools are more likely to label polished writing as copied. German courts separately ruled that the U.S. company Suno could not create music using works by members of Germany’s music copyright organization without their consent. Experts therefore argue that AI should remain a supporting instrument, not a substitute for human judgment.

“

AI tools should still remain an assistive means.