Databricks and Noma Secure Enterprise AI Inference Against Threats

Enterprise AI adoption is accelerating rapidly, but with it comes a critical security challenge: protecting AI inference, the stage where live models interact with real-world data. This phase is notoriously vulnerable to attacks such as prompt injection, data leaks, and model jailbreaks, which can expose sensitive information or compromise AI behavior. Recognizing this, Databricks Ventures and Noma Security have joined forces, backed by a $32 million Series A funding round, to deliver integrated, real-time security solutions tailored for AI inference.

Niv Braun, CEO of Noma Security, highlights that security concerns are the primary reason enterprises hesitate to scale AI deployments. Their joint approach with Databricks embeds advanced threat analytics and runtime protections directly into enterprise workflows, enabling organizations to accelerate AI adoption confidently and securely. This real-time defense is crucial because traditional cybersecurity measures often overlook the unique vulnerabilities present during AI inference.

Why AI Inference Security Demands Real-Time Protection

Gartner’s research underscores the urgency of securing AI inference, predicting that over 80% of unauthorized AI incidents through 2026 will stem from internal misuse rather than external attacks. This shifts the focus to integrated governance and continuous monitoring within AI workflows. Noma’s solution provides runtime defense with multilayered detectors, advanced natural language processing models, and proactive red teaming — a method of simulating attacks pre-production to uncover vulnerabilities early.

This proactive red teaming is vital for maintaining AI integrity from day one. By simulating sophisticated adversarial attacks before deployment, enterprises can identify and remediate security gaps, reducing time to production without compromising safety. Braun emphasizes that their runtime protections evolve alongside increasingly complex AI models, ensuring comprehensive security at every inference step.

Addressing Key AI Inference Threats with Databricks and Noma

The partnership targets several critical threat vectors at the inference stage:

• Prompt Injection: Malicious inputs that override model instructions, mitigated by multilayered prompt scanning and input validation.
• Sensitive Data Leakage: Real-time detection and masking prevent accidental exposure of confidential information.
• Model Jailbreaking: Runtime detection stops attempts to bypass safety mechanisms, ensuring appropriate AI outputs.
• Agent Tool Exploitation and Memory Poisoning: Continuous monitoring and integrity checks protect against unauthorized access and misinformation.

Leveraging Databricks Lakehouse for AI Governance and Compliance

Databricks’ Lakehouse architecture uniquely combines the governance strengths of data warehouses with the scalability of data lakes, centralizing analytics, machine learning, and AI workloads in a governed environment. This integration supports compliance with frameworks like OWASP, MITRE ATLAS, the EU AI Act, and ISO 42001, embedding transparency and regulatory adherence directly into AI operational workflows.

By aligning security controls with these standards, enterprises can confidently manage AI risk and governance throughout the AI lifecycle, from development to production inference.

Scaling Secure Enterprise AI with Integrated Threat Detection

As enterprises expand AI deployments, securing the inference stage becomes paramount. The Databricks-Noma partnership delivers a comprehensive solution combining governance, real-time threat analytics, and runtime protections. This approach addresses the top concerns of CISOs, enabling safe, compliant, and scalable AI adoption.

By embedding security directly into AI workflows and continuously monitoring for emerging threats, enterprises can confidently harness AI’s transformative potential without compromising safety or compliance.