"AI Models Wide Open to Cyberattacks, Analyst Warns"

According to Lasso Security researchers, while HuggingFace and GitHub developer platforms are important for developing Artificial Intelligence (AI) technologies, they also expose top-level organization accounts from Google, Meta, Microsoft, and VMware to threat actors. Lasso Security began its investigation in November, inspecting hundreds of Application Programming Interfaces (APIs) on the expertise-sharing platforms. Meta, the parent company of Facebook, was discovered to be especially vulnerable, with its Large Language Model Meta AI (LLaMA) exposed in many cases. According to the researchers, the investigation revealed a significant breach in the supply chain infrastructure, exposing high-profile Meta accounts. The consequences of this breach are far-reaching, as they successfully gained full access to Meta LLaMA 2, BigScience Workshop, and EleutherAI, with both read and write permissions. This article continues to discuss key findings from Lasso Security's investigation.

Cybernews reports "AI Models Wide Open to Cyberattacks, Analyst Warns"

Submitted by grigby1