Daxa Pebblo is an AI data security product delivered through a partnership between Daxa.ai and HPE, positioned as a "Secure AI Factory" solution for enterprises adopting Generative AI. It takes a shift-left approach to AI data security, intercepting and filtering sensitive information before it reaches the LLM as model context, rather than acting on outputs after exposure. The product is designed to address risks introduced by AI coding assistants and autonomous agents operating inside enterprise environments. These risks include leakage of secrets and intellectual property to external models, data overreach beyond role or project boundaries, prompt injection, supply chain attacks via third-party MCP servers, and unsupervised autonomous agent actions causing unintended system changes. Pebblo operates at the protocol level using MCP-native security, validating permissions and sanitizing payloads before they reach AI assistants. It enforces deterministic access controls at the data layer rather than relying on probabilistic prompt filtering. Core solution areas include: - Safe Agents: Protects enterprise data from AI coding assistants and embedded agents through policy-based guardrails and agent behavior controls. - Safe Inference: Reasoning-driven retrieval with anomaly detection to align context with user intent and enterprise policy. - Safe Connectors: Deterministic data filtering prior to LLM ingestion to keep sensitive or malicious data out of the AI pipeline. The product also provides real-time Data Loss Prevention, blocking secrets, credentials, and proprietary code from leaving the enterprise environment, and includes injection and supply chain defense capabilities for vetting third-party MCP servers.