The recently-formed Open Secure AI Alliance (Osaia) has developed a Shared AI Findings Exchange (Safe) proposal under the Linux Foundation to create a confidential incident-learning system for AI security findings and near misses.
With more than 120 member organisations, the alliance was built to pair openness with strong safeguards, clear rules against malicious misuse, rigorous evaluation and rapid remediation. The new proposal is an attempt to make AI systems easier to inspect and defend.
NVIDIA is contributing open models, model weights, data and new agent harness research to Osaia to speed the development of new cybersecurity tools and techniques. Its open source NVIDIA Labs Object-Oriented Agent (NOOA) project is now available on GitHub to make advanced AI safety capabilities more accessible for agent harnesses.
The alliance’s current work extends beyond models and harnesses. Members are developing guidance across identity, runtime controls, model scanning, privacy and secure coding, while the draft Safe framework would require members to report AI incidents and near misses through a staged notification process and publish remediation updates over time.
Broader ecosystem contributions, including Amazon’s Strands Agents and Cedar, and Visa’s open-sourced Visa Vulnerability Agentic Harness have led to a wider coalition of cloud, cybersecurity and enterprise software players trying to make AI defences more transparent and reusable.
Entelechy Asia 