AI Privacy and Misuse

Why AI Changes the Privacy Conversation

AI hasn't invented the idea of losing privacy, but it has changed its scale. Before machine learning became widespread, watching a single person closely took a person watching them. Today, models can scan thousands of hours of video, millions of records, or an entire person's online history in seconds. That shift, from privacy risk being expensive to being nearly free, is what makes AI a genuinely new kind of concern rather than just an old one with a new tool.

Surveillance and Facial Recognition

Facial recognition systems compare a face captured on camera against a database of known faces and return a likely match. Cities have used this to track suspects across networks of public cameras, some retailers use it to flag repeat shoplifters, and some employers use similar computer vision tools to monitor attendance or behavior. The core objection isn't the technology alone: people are often identified and tracked without being asked, the matching isn't always accurate (error rates have historically been higher for some demographic groups), and being watched constantly can change how people act in public, even when they've done nothing wrong.

• Public-space cameras paired with face-matching software used by some police departments
• Retail stores flagging repeat visitors using in-store cameras
• Employers monitoring keystrokes, screen activity, or location during work hours
• License-plate readers that log where and when a car was seen
• Apps and phones building a location history that reveals someone's daily routine

Deepfakes and Synthetic Media

A deepfake is a video, image, or audio clip generated by an AI model to convincingly show someone doing or saying something they never did. The model is trained on real examples of a person's face or voice and learns to generate new, fake content in that same style. This has legitimate creative uses, such as dubbing a film into another language while matching the actor's lip movements, but it has also been used for fraud, like cloning an executive's voice to authorize a fake wire transfer, and for creating non-consensual explicit images or fabricated political statements. A side effect, sometimes called the 'liar's dividend,' is that real footage can now be dismissed as fake, making it harder to trust any recording at all.

Data Scraping and Consent

Many AI models, especially large language and image models, are trained on data collected from the public internet: articles, forum posts, photos, code, and more. Much of this was published for people to read or view, not necessarily for a company to copy and use to train a commercial product. This has led to real disputes: artists and photographers whose work appears in image-generation training sets without payment or permission, writers and news organizations concerned about text used to train chatbots, and ordinary users whose social posts or reviews end up as training examples. There's no fully settled answer yet to where 'publicly visible' ends and 'consented to reuse' begins.