A police officer used AI and official databases to create deepfakes of 24 women: abuse of access is the problem

By Inquirer.com · November 2, 2026.

Stephen Kamnik, 40, a former Pennsylvania State Police corporal, was sentenced on Monday to five to ten years in state prison for using AI to create thousands of fake pornographic images (deepfakes) of women, The Philadelphia Inquirer reports. Montgomery County Judge Todd Eisenberg also designated him a sexually violent predator: he must register as a sex offender for life and is barred from working in security, law enforcement or IT.

According to the judge, the victims are about two dozen women, including co-workers and strangers Kamnik encountered in the course of his police work. According to prosecutors, quoted by the paper, he generated the images from photos and videos recorded without their consent and from photographs in official state databases he had access to through his position. Among the files, investigators found child sexual abuse material, including images depicting a 12-year-old girl. According to prosecutors, he tried to delete part of that collection after being discovered.

Kamnik pleaded guilty in April to offenses including child sexual abuse, unlawful use of a computer and official oppression (abuse of authority). In a tearful statement, he took responsibility for his actions. His lawyer asked for a long probation with treatment; Assistant District Attorney Kimberly Moraski argued for incarceration to protect the public. The judge stressed that citizens trust the police to protect them and that Kamnik used his position "to find people."

Our take: this case is not about a new technology, but about an old equation with a multiplier. Abuse of personal data by those with privileged access is nothing new; what has changed is that generative AI turns an ID photo into fake sexual material in minutes, with no technical skill required. The cost of causing this harm has collapsed, and the criminal justice system already treats the creation of this material as a serious crime in itself, which is a good sign.

The thesis we advocate is that defense must act on two fronts. The first is access: official databases containing faces and identities are a sensitive asset, and they need audit logs, least privilege and detection of anomalous queries, because the insider with legitimate credentials is the hardest risk to spot. The second is the generation layer: models and services that make it possible to undress or sexualize real people, and especially minors, must be blocked or held accountable.

In general, the short term in this area is uncomfortable: more impersonation, more extortion and more harm to victims, mostly women, while laws and platforms lag behind. In the long term, I still believe the same technology can deliver abundance and better societies, but only if consent and the safeguarding of identity are treated as critical infrastructure rather than an afterthought. This case is a reminder that progress is also measured by how we protect those on the wrong end of the camera.