Deepfake executives are infiltrating video meetings as researchers develop a real-time detection system
Researchers are developing a real-time system to detect deepfake executives during video meetings and prevent corporate fraud.
Video conferencing has become an essential part of modern business, but security experts are warning that it is also becoming a new target for cybercriminals. As artificial intelligence continues to improve, attackers are increasingly using deepfake technology to impersonate company executives and employees during online meetings, making fraud more convincing than ever.
Table Of Content
Researchers at the Fraunhofer Institute for Secure Information Technology (SIT) are developing a real-time warning system to detect potential deepfake activity during corporate video conferences. The technology aims to alert participants while a meeting is still in progress, allowing employees to verify suspicious requests before acting on instructions that could result in major financial losses.
Deepfake technology creates new risks for businesses
Cybercriminals have long relied on phishing emails and fake phone calls to trick employees, but experts say video meetings are becoming an attractive new avenue for fraud. The combination of familiar faces, recognisable voices and the pressure of a live meeting can make it much harder for people to question unusual requests.
According to the researchers, attackers are exploiting the trust that naturally exists between colleagues. By using AI-generated video and voice technology, criminals can imitate senior executives or team members with increasing accuracy, making fraudulent instructions appear genuine. The proposed warning system is intended to detect signs of manipulation before an employee authorises a payment or shares sensitive company information.
The growing sophistication of deepfake technology means the warning signs that once made fake videos obvious are becoming less noticeable. Earlier versions often displayed poorly synchronised facial movements or unnatural expressions, but newer AI models can produce far more realistic results, making visual inspection alone an unreliable defence.
A widely reported incident in Hong Kong demonstrated how convincing these attacks have become. An employee joined what appeared to be a routine video conference with several colleagues, only to discover afterwards that every other participant had been generated using artificial intelligence. Investigators said the criminals had obtained internal company video material and created AI-generated voices to impersonate employees. During the meeting, they convinced the victim to transfer nearly €24 million before the deception was uncovered when the victim spoke with the real manager.
Research highlights how easily people can be deceived
The Hong Kong case illustrates that even experienced employees can struggle to recognise sophisticated deepfake attacks, particularly when several familiar faces appear together in the same meeting. Seeing multiple trusted colleagues can reinforce the impression that a conversation is legitimate, reducing the likelihood that participants will question unexpected requests.
Academic research has reached similar conclusions. A 2025 study examined how well students could identify a manipulated video featuring their professor. Researchers used a publicly available real-time face-swapping tool to create the altered footage. Out of 34 participants, 25, representing 74 per cent of the group, failed to detect that the video had been manipulated before the researchers informed them.
The findings suggest that human judgement alone may no longer provide sufficient protection against AI-generated deception. As deepfake technology becomes more accessible and realistic, organisations may need to introduce additional safeguards rather than relying solely on employee awareness or security training.
Security specialists believe attackers will continue refining their methods, combining convincing visual impersonation with psychological pressure. Fraudsters often create a sense of urgency by requesting immediate financial transfers or confidential information, making it more difficult for employees to pause and verify the request’s authenticity.
Real-time alerts could strengthen corporate security
Fraunhofer’s detection system builds on the institute’s existing media forensics research, which combines deep learning with advanced signal processing techniques. Instead of focusing on obvious visual flaws, the system analyses subtle indicators such as blurred facial structures, inconsistencies across different image regions and other small irregularities that may reveal AI-generated content.
The goal is not to replace human decision-making but to provide an early warning when something appears suspicious. A timely alert could encourage participants to pause the meeting, contact colleagues through another communication channel or seek additional confirmation before following potentially fraudulent instructions.
However, researchers acknowledge that deepfake detection technology is not without limitations. Martin Steinebach, a researcher involved in the work, has previously warned that automated deepfake detection systems can produce false positives and false negatives during live video calls. Incorrect warnings could confuse users or undermine confidence in the technology if alerts are triggered too frequently.
Despite these challenges, experts believe real-time detection tools could become an important layer of defence as AI-powered impersonation attacks continue to evolve. Rather than relying on a single security measure, businesses are increasingly expected to combine technical detection systems with employee verification procedures and stronger internal controls. Together, these measures may help reduce the risk of costly fraud as deepfake technology becomes an increasingly common threat in corporate communications.
