Page 181 - CyberSecurityEssentialsEnglish

4.9 Disinformation and Synthetic Content: When Seeing and Hearing Are Not Enough
◆ Key idea: the realism of a medium is not evidence about its source.
Deepfakes and voice cloning can produce a convincing image, video, or voice. Do not rely
on an odd appearance or a single visual glitch; the tools keep improving. Check the
channel, the context, and the source, make contact through a known route, and do not
take a financial or sensitive decision under the pressure of urgency.

Misinformation is false content that is sometimes spread without intent to deceive, while
disinformation is false or manipulated content spread deliberately to deceive. Content may

also combine a true fact, an old image, and a new caption, so not every piece of
disinformation is fully generated. What AI changes is the cost of producing many tailored
copies.

A deepfake includes synthetic or altered media that imitates a person or an event. Voice

cloning may use a short published clip to build a convincing voice. Consumer protection
authorities have warned about the use of these techniques to impersonate individuals and

executives and to request transfers of money or information [11][12].

Behavioral warning signs matter more than technical flaws: urgency, secrecy, a request to
bypass a procedure, an unusual transfer, sending a verification code, changing a bank
account, or emotional pressure. Even if the voice is real, the account may be compromised,

so the request must be verified, not only the voice.

Use out-of-band verification: end the call and dial a known number, start a new message
from an official directory, or ask another person. A family or a team can agree on a
verification word or a dual procedure for payments, but the word must never be placed in a

public channel.

Provenance data helps to establish when content was created and modified. Content
provenance standards provide a way to bind that data to the content cryptographically, and
it may include the tool, the time, and the edits [13]. Later revisions of the specification have

added formats and clarifications [21]. A related technique is watermarking, in which a
marker is embedded in the content itself, either visibly or in a way only software can read.
Both approaches share the same weakness: a marker can be stripped by re-encoding,

cropping, or a screenshot, and its absence proves nothing. Provenance also does not mean
the content is truthful; it offers one signal of origin within a body of evidence.

180