Open Secure AI Alliance Unveils Agentic AI Cybersecurity Guidelines

The Open Secure AI Alliance has unveiled new guidelines for agentic AI cybersecurity, with over 120 member organizations contributing to the effort. The Linux Foundation recently issued a Request for Comments on its proposed Shared AI Findings Exchange (SAFE) guidelines, which aim to transform individual agentic cybersecurity incidents into shared protective measures for the entire AI ecosystem.

NVIDIA is making significant contributions across the full AI security stack, including the NVIDIA Labs Object-Oriented Agent (NOOA) research harness and the NVIDIA OpenShell runtime. The company's open model families, such as Nemotron for agentic AI, ship with open weights, datasets, and training techniques to enhance transparency and inspectability.

Other alliance members are also contributing extensively to the defensive stack, including Okta, Palo Alto Networks, Red Hat, Amazon, Capital One, Cloudflare, Microsoft AI Red Team, Wiz, and Visa. These contributions include tools for identity, harnesses, models, and resilience, as well as specialized AI models built for defense.