Responsibility, Law & Governance
Deepfake
Faces, voices, scenes — synthetically generated and barely distinguishable from the real thing. The consequences are concrete: the CEO’s voice on the phone, forged evidence videos, fraud by video call. The flip side is just as dangerous: even genuine material can be dismissed as “surely fake.” Processes help where eyeballing does not — callback routes, code words, verification through a second channel.
An example
The case that matters in practice for companies is not the faked politician video but the phone call: a cloned voice — thirty seconds of material from a public talk are enough — instructs the accounting department to make an urgent transfer. A callback to the known number would have exposed it. The instruction to do exactly that did not exist, because nobody had anticipated the case.
A closer look
Image, voice or video is generated so that it appears to show a real person. A few seconds of material are by now enough for a convincing voice forgery, and the cost has fallen to practically zero.
Why it matters
The effective response is organizational, not technical: procedures that do not depend on a voice or a face being real. A callback to a registered number for payment instructions, a four-eyes principle above a threshold, an agreed code word for urgent exceptions. This costs nothing and works immediately. Detection software, by contrast, is a race that the forgers win, because every detection serves as a training signal for better forgeries.
Commonly misunderstood
The most widespread use is not political disinformation but everyday fraud: the call from the supposed managing director. What helps against it is a callback procedure, not detection software.