NEWS

The emergence of self-augmenting malware utilizing Large Language Models (LLMs) poses a significant threat to cybersecurity, as highlighted in a recent report by Recorded Future.

By leveraging generative AI, threat actors can modify existing malware code to evade detection by string-based YARA rules, thus lowering detection rates. This capability enables attackers to maintain the original functionality of the malware while ensuring syntactical correctness in the generated code.

While this technique offers a means to bypass conventional detection methods, it does have limitations, primarily related to the amount of text an LLM can process at once, making it challenging to operate on larger code bases effectively.

Beyond evading detection, AI tools can be exploited for various malicious purposes, including the creation of deepfakes to impersonate individuals for influence operations and reconnaissance on critical infrastructure. By utilizing multimodal models, threat actors can extract metadata from public images and videos to gather strategic information for follow-on attacks.

Recent warnings from Microsoft and OpenAI indicate that threat actors, such as APT28, are already leveraging LLMs to acquire in-depth knowledge of satellite communication protocols and radar imaging technologies, emphasizing the importance of organizations scrutinizing publicly accessible imagery of sensitive equipment and taking appropriate measures to mitigate risks.

Moreover, researchers have demonstrated the potential for malicious actors to exploit LLM-powered tools by jailbreaking them and producing harmful content using ASCII art inputs. This attack, known as ArtPrompt, exploits LLMs’ poor performance in recognizing ASCII art to bypass safety measures and elicit undesired behaviors.

In summary, the increasing sophistication of AI-driven malware presents a pressing challenge for cybersecurity, requiring proactive measures from organizations to defend against emerging threats and mitigate risks associated with AI exploitation.