Shulou(Shulou.com)12/24 Report--

CTOnews.com, December 8, according to the official Meta press release, Meta yesterday launched an AI inspection suite called Purple Llama, which claims to provide a range of tools to help developers test their own AI security and "responsibly deploy generated AI models and applications."

CTOnews.com has learned that the Purple Llama suite currently provides "CyberSec Eval" evaluation tools and Llama Guard "input / output security classifiers", and Meta claims that more tools will be added to the suite later.

According to Meta, the Purple Llama suite, which aims to standardize its own Llama language model and can also work on other competitors'AI models, is based on the "purple team concept" in the security field.

Meta claims that the Purple team is a combination of "attacking the red team and defending the blue team", and that the role of the Purple Llama suite is to assess and mitigate the potential threats of the AI model and to "strike a balance between attack and defense strategies."

▲ source Meta specifically, the Purple Llama suite can currently quantify the indicators of network security risks of large models, assess possible code vulnerabilities, and provide insights into the security of the model. In addition, it can also check and filter the input and output of the model to prevent hackers from "jailbreaking" related models.

Meta said that in the future, the company will continue to test the security of the AI model. In addition to continuously improving the Purple Llama suite, Meta will also work with Microsoft, AMD, Amazon, Google, Intel, Nvidia and other companies to integrate the Purple Llama evaluation suite into the industry model evaluation benchmark.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.