SophosAI speaking at Virus Bulletin ’24: Leveraging multimodal AI as a “sensory extension” for cybersecurity
During this week’s Virus Bulletin 2024 event in Dublin, Ireland, the principal data scientist at Sophos, Younghoo Lee, is set to deliver a session on the research conducted by SophosAI in integrating sophisticated language model (LLM) artificial intelligence into the intricate task of classifying email and web content. Lee’s talk will delve into the deployment of “multimodal” AI—which involves analyzing both the text and visuals in email and web content to categorize the content as either benign, undesired, or harmful.
In his paper titled, “Multimodal AI: Enhancing Cyber Defense with a Sixth Sense,” Lee sheds light on the fact that threat actors can exploit generative AI to craft persuasive phishing emails and websites, as evidenced by SophosAI’s “Scampaign” research unveiled last year. Moreover, conventional machine learning methodologies for phishing and spam detection, which rely on training models with known malicious and unwanted (“spam”) content, may overlook novel strategies that deviate from existing patterns.
Nevertheless, the very technology explored by SophosAI as a potential conduit for malicious email and web content could also be leveraged to identify such content, in addition to categorizing new websites lacking established reputation data. Through the implementation of multimodal AI that scrutinizes all facets of content using LLMs––models trained on extensive datasets of online content––to inspect email headers, content, logos, and visuals, it becomes feasible to detect fresh instances of phishing emails and other malicious or undesired content without the need for specific training data.
Lee’s session is scheduled for 11:30 AM local time on Friday, October 4. An elaborate article on the research will be made available subsequent to the session.
About Author
