Cybercrime

"With access to general purpose AI models, such as OpenAI’s GPT-4 underlying ChatGPT, malicious actors are able to produce a higher quality of fake content – for example texts and media – faster.145 While these models could also be used to target IT systems (“cyber crime in the narrow sense”), for example, through phishing emails or assisting in programming malicious software,146 it is not yet clear how strong the impact of this technology will be here."(p. 28)

"For the initial element, the attack method, general purpose AI models can generate persuasive and personalised content that is often more convincing than traditional fraudulent communication.150 These models can also be utilised by criminals to faster and more efficiently set up their infrastructure, the second element of the criminal activities. General purpose AI models are able to authentically imitate the style or rhetoric of a person or organization, increasing the credibility of the communication and thus the effectiveness of the criminal activity."(p. 29)

"Additionally, OpenAI’s ChatGPT has already been used for assisting in programming malicious software that may be used in criminal activities targeting IT systems.156 Due to many aspects, such as an already existing low-cost supply of malware for this purpose, it is not yet clear how strong the impact of general purpose AI models will be in this area. General purpose AI models are already and will increasingly be part of the diverse toolbox of cyber criminals."(p. 30)