Cisco, Meta push open-source AI for cybersecurity

Open-source large language models are rapidly redefining cybersecurity as startups and tech giants leverage their speed and adaptability to counteract machine-speed threats, VentureBeat reports.

At RSAC 2025, Cisco, Meta, and ProjectDiscovery revealed significant open-source initiatives that signal a collective push toward unified and community-driven security infrastructures. Cisco unveiled Foundation-sec-8B, an 8-billion-parameter LLM built for cybersecurity, trained on curated threat datasets and released under the Apache 2.0 license for flexible enterprise use. “These are real-life examples,” Cisco’s Jeetu Patel said, stressing collaboration over competition. Meta expanded its AI Defenders Suite with Llama Guard 4 and LlamaFirewall to detect policy violations, prompt injections, and agent misalignments. Its new benchmark suite, CyberSec Eval 4, tests AI performance in real-world SOC conditions. Meanwhile, ProjectDiscovery’s open-source vulnerability scanner, Nuclei, earned the “Most Innovative Startup” award for its community-driven model and rapid CVE detection capabilities. Together, these developments emphasize how open-source LLMs are becoming essential infrastructure for scalable, cost-effective, and highly specialized cybersecurity defenses.