The advantages that chatgpt brings are multiple, since they can write lines of code in several programming and scripting languages when given an entry in natural language.
On this, Sofos has published the report “The application of linguistic processing to cyberdefensas”, on how the cybersecurity sector can take advantage of GPT-3, the language model used by chatgpt, as a co-pilot to help defeat cyberbots.
“ANDn Sofos, we have long seen AI as an ally rather than as an enemy, which makes it a fundamental technology for sophos, and GPT-3 is no different. The cybersecurity sector should pay attention not only to potential risks, but also to the opportunities offered by GPT-3«, Explains Sean Gallagher, principal researcher of threats of sophos.
SOFHOS X-PE researchers have been working on three experimental projects:
- Project 1: Consultation interface in natural language to seek malicious activity. The first application that Sophos has tried with the “Few-Shot Learning” learning method is an interface of consultation in natural language to filter malicious activity in XDR telemetry. The widest model created by Sofos for this project has given precise answers in more than 80% of the cases when dealing with natural language questions that used data they had seen as part of the training, and in 70.5% of the time when dealing with questions that included data that the model had not seen before.
Sopos demonstrates how to convert the AI of Chatgpt into an ally against cybercrime
- Project 2: GPT -based spam detector. Machine Learning has already been applied prior to spam detection using different types of model. However, Sofos has tried a new spam filter using chatgpt and discovered that, compared to other models, the GPT-3 filter is significantly more precise and far exceeds other traditional Machine Learning methods when the amount of data used for training is small.
- Project 3: Tool for analyzing command lines executed in binary files “Living of the Land” (Lolbin). Finally, Sofos researchers have created a program to simplify the inverse engineering process of Lolbins’ command lines. For developers and experts, this type of reverse engineering is difficult to perform, since it includes a lot of confusion and are long and difficult codes to analyze; But it is also essential to do so to understand the behavior of Lolbins and stop these types of attacks.
