AI fakes: trust starts with the right question
Deceptively real images, voices and texts make first impressions unreliable. Protection comes not from distrusting everything, but from verifiable sources, context and clear labelling.
A video shows a familiar face recommending an investment. A voice on the phone sounds like your daughter. An image looks like proof. With generative AI, such deception is becoming cheaper and more convincing. But that does not mean no information can be trusted any more. The more useful question is: which steps make a claim verifiable?
Not every artificial image is a fake
AI-generated content can be harmless, creative or clearly labelled. It becomes problematic when it creates a false impression about origin, statement or consent. The Federal Office for Cybersecurity warns about fake interviews with well-known figures used for investment fraud. Face swapping can also be misused for blackmail or bullying.
The harm often arises not from the technology alone, but from the interplay of emotional urgency, familiar faces and rapid forwarding. Anyone being urged to share or pay into an alleged investment immediately should not try to expose the video with the naked eye alone. It is more useful to check the claimed source through its official channel, look for independent reports and take a pause before making financial decisions.
Origin counts for more than a gut feeling
Verification software can provide clues, but no final guarantee of truth. Watermarks are sometimes missing, metadata can be lost, and detectors develop more slowly than new methods. Origin therefore takes priority: who published something first? Is the account verified? Is there a primary source, a complete video or an official statement? Do place, time and context fit?
For newsrooms, the same principle applies in professional form: secure original files, confirm sources independently, do not adopt statements from a viral clip alone, and visibly label synthetic material. In its work on AI governance in journalism, OFCOM notes that detection tools can be useful, but that the technology keeps evolving.
Switzerland is building rules – and still needs judgement
The federal government has announced that it will prepare a preliminary draft for AI regulation by early 2027. For AI-generated or altered content of public interest, it also points to the transparency requirements of the European AI Act. At the same time, data protection, personality rights and criminal law provisions already apply in Switzerland. The law is therefore not empty; it must be applied and supplemented for new forms of deception.
This matters in practice, especially for schools, families and older people: a good rule only helps once people know where to report items, how to document fraud attempts and which authority is responsible. Media literacy is not specialist knowledge for technology fans, but an everyday skill.
Trust in the digital sphere does not arise from treating every image as suspicious. It arises when claims have a traceable origin, deception is penalised, and users know how to pause and check. For Switzerland, that is a democratic infrastructure task – as much as cybersecurity and independent journalism.
