Meta Debuts Security Tools in Llama AI Model

Meta has launched “Purple Llama,” a project tied to its Llama artificial intelligence (AI) tool.

Announced Thursday (Dec. 6), Purple Llama centers around safety tools and evaluations to help developers build responsibly with their AI models.

“Why purple? Borrowing a concept from the cybersecurity world, we believe that to truly mitigate the challenges that generative AI presents, we need to take both attack (red team) and defensive (blue team) postures,” the tech giant said on its blog. “Purple teaming, composed of both red and blue team responsibilities, is a collaborative approach to evaluating and mitigating potential risks.”

Meta unveiled its Llama 2 in July in partnership with Microsoft. The company said Thursday its Llama models have been downloaded more than 100 million times. 

According to the blog, components of the project will be licensed “permissively,” allowing for both research and commercial use to foster collaboration and standardize trust and safety tools.

In addition, the company said it is sharing what could be the first industry-wide set of cyber security safety evaluations for large language models (LLMs). 

“These benchmarks are based on industry guidance and standards and are built in collaboration with our security experts,” Meta said, and designed to provide tools to address risks outlined in the White House’s AI safety commitments. 

These tools include metrics for quantifying LLM cybersecurity risk, ways to gauge the frequency of insecure code suggestions, and tools to make it difficult for LLMs to generate malicious code or help carry out cyber attacks.

“We believe these tools will reduce the frequency of insecure AI-generated code suggested by LLMs and reduce the helpfulness of LLMs to cyber adversaries,” Meta said.

The news comes one day after reports that Meta was testing more than 20 generative AI features across its platforms: Facebook, Instagram, Messenger and WhatsApp.

These features cover aspects of social media that include search, ads and business messaging, Bloomberg News reported.

Ahmad Al-Dahle, Meta’s vice president of generative AI, told Bloomberg that the company is committed to creating a better community, helping users express themselves and build more useful products.

“That generally shows up in the metrics if we do a good job,” Al-Dahle said. “We see more usage, positive feedback, people having a good time.”

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.