The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Databricks announced its founding membership in the Open Secure AI Alliance, a coalition of over 75 organizations including NVIDIA, aimed at advancing AI safety and security through open research and shared tools. The alliance focuses on three core areas: securing AI systems against emerging threats, using AI to enhance cyber defense, and ensuring transparency across the AI agent stack, from models to governance layers. Members contribute open models, harnesses, tooling, and frameworks to accelerate the development of new security techniques and mitigate risks in enterprise environments.

The alliance emphasizes the need for an open execution stack to secure AI agents, which are complex systems comprising models, harnesses, and guardrails. Databricks highlighted its role in providing governed data, agent capabilities, and open tools for security, governance, red teaming, and cyber defense. The company introduced Omnigent, an open-source meta-harness that enforces policies, spend limits, and sandbox isolation across multiple agent frameworks, supporting NVIDIA’s OpenShell for layered protection and auditable governance.

Databricks also released the Databricks AI Security Framework (DASF) 3.0, a vendor-agnostic framework cataloging 97 technical security risks across 13 components and mapping them to 73 mitigation controls. DASF 3.0 includes dedicated coverage for agentic AI threats such as memory poisoning and goal manipulation, providing a blueprint for defenders to transition from risk identification to mitigation using open, cross-referenced standards like MITRE ATLAS and NIST.

Additionally, Databricks introduced BlackIce, an open-source red-teaming toolkit that bundles 14 AI security tools into a single environment mapped to MITRE ATLAS and DASF. The toolkit enables security teams to test systems against threats like prompt injection and supply-chain attacks without requiring extensive manual setup. Databricks’ Security Lakehouse, powered by Lakewatch, further supports AI-enabled cyber defense by unifying security data in open formats, allowing teams to deploy AI agents for threat detection and response at scale while avoiding vendor lock-in.