Reviewing Malware with LLMs: OpenAI vs. Vertex AI

Large Language Models (LLMs) are evaluated for their ability to assess the risk of potentially malicious code snippets, with a focus on the comparison between OpenAI's models and Google's Vertex AI. The assessment process involves using LLMs to assign a risk score on a scale of 0 to 9, rather than a binary classification, which helps in more nuanced evaluations. Improvements in the LLM-assisted review process include removing comments to reduce the risk of prompt injection and expanding context size. In the comparison, OpenAI's gpt-3.5-turbo and Google's text-bison often agree on risk scores, but diverge in cases where code is obfuscated. OpenAI's gpt-4 model is noted to provide superior explanations and risk evaluations for non-obfuscated code. The blog also discusses the manageable risk of prompt injection in this context, due to the requirement for code to adhere to syntactic rules, which allows for pre-processing steps like removing comments to mitigate risks.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.