SecurityWeek argues for in-call deepfake detection with help desk breaches it never ties to AI

SecurityWeek says there is little empirical evidence that awareness training works against social engineering ^[15]. It says humans cannot be expected to spot manipulation built for their psychology, and that AI deep fakery is making the tricks harder to see ^[16]. The AI is in the argument. The incident accounts describe callers posing as employees, and a caller talking a staff member through an Entra step ^[1][5][13].

Netarx's checks come in two kinds. One looks for synthetic media. In cloned voices that means unnatural background silencing and compression artifacts. On video it means lip movements that do not match the audio, unnatural blinking and warping around hairlines ^[8][9]. The other kind tests whether the device on the call belongs to its authorized user, using device fingerprints, EXIF data, compression signatures and location mismatches ^[8].

In my view only the device checks apply to the breaches SecurityWeek cites. I'd also expect the first fix in both cases to be procedural: limiting what help desk staff may reset or approve for a caller, whatever the caller sounds like. In the Brinks Home case the employee completed the authentication step on the attacker's behalf ^[5].

No single signal decides the verdict ^[10]. The combined signals drive a light on screen during the call. Green means identity and device are verified. Amber means the source is unknown or the metadata looks wrong. Red means synthetic media was found ^[11]. SecurityWeek calls automatic blocking extreme, so the employee decides whether to disengage ^[11]. The casino attackers rang in asking for password resets ^[1]. I'd expect staff to act on amber only if the genuine employees who call with the same request rarely trigger it. The article does not report detection or false-positive rates ^[14].

SecurityWeek says use of this kind of detection is increasing and improving, and it names one product, Netarx ^[18]. It calls Netarx an early case of technology supplanting human detection and says it will not be the last ^[17]. Netarx's NIK runs on Windows, macOS, Chrome, iOS and Android. The company also publishes an Impact Database of incidents where human deception was decisive ^[12].