Saturday 22 March 2025
Phishing attacks have become increasingly sophisticated, exploiting human vulnerabilities through social engineering tactics to deceive individuals into revealing sensitive information. Traditional detection methods often fail to identify new or cleverly disguised phishing attempts due to their reliance on known patterns and technical indicators.
To combat this issue, researchers have developed a conversational AI-powered assistant called Cyri. This innovative tool is designed to support human users in detecting and analyzing phishing emails by leveraging large language models (LLMs). Cyri scrutinizes emails for semantic features used in phishing attacks, such as urgency, authority, impersonation, exclusivity, and undesirable consequences.
One of the key challenges in developing Cyri was creating a comprehensive dataset of phishing emails. The researchers generated 420 phishing emails, with 20 emails dedicated to each of the 21 identified semantic features. This diverse set of emails allowed them to test Cyri’s ability to detect phishing attempts across a wide range of topics.
The researchers evaluated Cyri’s performance using a combination of human evaluation and automated testing. They found that Cyri achieved high accuracy rates, with an overall detection rate of 95.24%. Moreover, the tool was able to provide clear explanations for its detections, making it easier for users to understand why an email is potentially malicious.
Cyri has several features that make it more effective than traditional phishing detection methods. For example, it can detect emails that use emotional appeals or try to create a sense of urgency. It also provides users with information about the likelihood of an email being legitimate, which helps them make informed decisions.
The researchers tested Cyri on a group of 10 participants, including both experts and non-experts in computer security. The results showed that Cyri was highly effective in helping users identify phishing emails, regardless of their level of expertise. Participants reported feeling more confident in their ability to detect phishing attempts after using Cyri.
Cyri’s success can be attributed to its ability to provide clear explanations for its detections. This feature helps users understand why an email is potentially malicious and what steps they should take to protect themselves. Additionally, the tool’s conversational interface makes it easy to use and intuitive.
The development of Cyri has significant implications for phishing detection and management. It provides a new approach to detecting phishing emails that is more effective than traditional methods. Moreover, its ability to provide clear explanations and help users understand why an email is potentially malicious can improve user awareness and education about phishing attacks.
Cite this article: “Conversational AI-Powered Assistant Detects Phishing Attacks with High Accuracy”, The Science Archive, 2025.
Phishing, Detection, Ai-Powered, Conversational, Semantic Features, Language Models, Email Analysis, Phishing Attacks, Human Vulnerability, Cybersecurity
