SecuPi launches data security platform for self-hosted AI

SecuPi unveiled a data security platform designed to protect sensitive information in self-hosted AI environments across on-premises, private cloud, VPC and air-gapped deployments. The platform is aimed at regulated industries that need AI access controls, monitoring and policy enforcement without moving data or controls outside customer infrastructure.

Why it matters: - Banks, healthcare providers, government agencies and other regulated enterprises are moving to self-hosted AI to keep sensitive data and infrastructure under their own control. - SecuPi is targeting the gap between hosting an AI model internally and actually controlling what the model, users and agents can access, see and do. - The platform is meant to support privacy, compliance and data-sovereignty requirements in environments where exposure of sensitive information carries operational and regulatory risk.

What happened: - SecuPi announced its Data Security Platform for self-hosted AI environments. - The platform is built to protect sensitive data from preparation and retrieval through inference and agent actions. - SecuPi said the system keeps data and security controls inside customer-controlled infrastructure. - CEO and co-founder Alon Rosenthal said keeping an AI model inside an organization’s environment is only the first step, and that runtime access control, encryption, privacy protection and real-time auditability are also needed. - Rosenthal also said the platform can operate in an air-gapped environment.

The details: - The platform combines runtime attribute-based access control, real-time monitoring of sensitive-data activity, a kill switch for malicious agent activity, and field-level protection using format-preserving encryption, tokenization and dynamic masking. - SecuPi enforces data access policies in real time across databases, data lakes, files, applications and analytics platforms. - The system applies fine-grained authorization at the object, table, row, column and field levels, depending on the data source. - The company says the controls help ensure AI workloads receive only information authorized for a legitimate business purpose. - Field-level protection can reduce sensitive-data exposure while preserving data formats where needed. - AI data preparation features can discover and automatically de-identify sensitive information before training, retrieval, inference or analytics. - Real-time monitoring can analyze behavior and block agent activity that violates policy. - Identity and auditability features link agent activity to the human on whose behalf it is performed and maintain tamper-resistant records of access and actions. - Centralized policies can apply consistent controls across structured and unstructured data in supported enterprise systems. - The platform is designed to work alongside self-hosted AI models in on-premises data centers, customer-controlled private clouds, virtual private clouds and air-gapped or “dark-room” environments. - Organizations can apply consistent security policies within approved geographic and operational boundaries. - SecuPi says supported integration points include transparent instrumentation agents, database and API gateways, APIs, SDKs, user-defined functions and command-line tools. - Supported connectivity spans Model Context Protocol servers, Spark, Kafka, NiFi and Python pipelines, plus Hive, PostgreSQL, MongoDB and Trino environments. - The architecture is intended to protect AI workflows including data preparation, retrieval-augmented generation, inference and agent actions. - SecuPi says the platform complements existing data catalogs, identity providers, security tools and AI governance platforms by translating governance context and access policies into runtime enforcement. - SecuPi also said the platform serves organizations in financial services, healthcare, telecommunications, manufacturing and other data-sensitive industries. - The company said it supports on-premises, customer-controlled cloud and air-gapped deployments. - SecuPi said it is recognized as a Gartner Cool Vendor. - The release directs readers to more information. - The release also includes SecuPi’s LinkedIn page at the company’s social profile.

Between the lines: - Self-hosted AI reduces some exposure, but it does not automatically solve access control, insider risk or agent behavior problems. - SecuPi is positioning runtime enforcement as the missing layer between AI deployment and AI governance. - The emphasis on air-gapped and dark-room support suggests the company is aiming at customers with the strictest isolation requirements.

What's next: - Organizations adopting self-hosted AI can evaluate whether SecuPi’s controls fit existing infrastructure and governance workflows. - The platform’s practical impact will depend on how well it integrates with enterprise data systems, AI tools and security stacks already in place. - Regulated buyers will likely focus on whether the platform can enforce policy without slowing AI access or creating operational friction.