The Linux Foundation shared a Request for Comments on the Shared AI Findings Exchange guidelines to convert agentic cybersecurity incidents into ecosystem-wide protections, NVIDIA reported today.
A working group within the Open Secure AI Alliance is drafting the guidelines, known as SAFE, as the annual Black Hat conference opens in Las Vegas. The alliance now includes more than 120 member organizations, with NVIDIA, Cisco, CrowdStrike, Hugging Face, and Red Hat contributing to the initial proposal.
Alliance members plan to confidentially submit and analyze AI security incidents and near misses under the proposal. The framework outlines procedures to inform impacted parties, identify recurring control failures, and publish evidence-based recommendations to reduce systemic risk.
Defensive tools and software
Alongside the proposed guidelines, NVIDIA detailed open-source security tools across its stack. These include the NVIDIA Labs Object-Oriented Agent research harness on GitHub, the OpenShell runtime, and Garak, an open-source vulnerability scanner for large language models.
Amazon and Visa joined the Open Secure AI Alliance today as its newest members. Amazon contributed its Strands Agents toolkit and Cedar authorization language, while Visa contributed its Vulnerability Agentic Harness to assist with flaw remediation.
Other member organizations released additional defensive tools across the security stack. Cisco introduced DefenseClaw for runtime governance on top of OpenShell, Mistral released its Shieldstral multimodal safety classifier model under Apache 2.0, and Uber shared components of its Agentic AI Detection and Response system, which currently processes over 200,000 agent sessions daily across 30,000 endpoints.
Alliance members gather today, Tuesday, Aug. 4, at 5:15 p.m. PT for a group photo outside the Main Stage in the Business Hall at the Mandalay Bay Convention Center.
