Concentric AI Semantic Intelligence

Enforces document-level access at retrieval time so a user receives only context they are authorized to see, filtering before the vector search, after retrieval, or both.

Intercepts prompts and completions to prevent sensitive data (PII, credentials, internal IP), from being transmitted to external LLM services or returned in model responses.

Records prompts, completions, and metadata for all AI interactions with tamper-resistant storage, supporting compliance, forensics, and policy investigation.

Discovers and classifies sensitive data (PII, PHI, payment data, IP, secrets) across structured and unstructured stores by combining deterministic techniques such as patterns, keywords, and validators with AI/ML techniques such as unsupervised clustering and small language models. Breadth of the technique blend, and whether classification extends to prompts, model outputs, and vector databases, are the primary differentiators; products that rely on pattern matching alone sit at the low end.

Maps effective permissions to sensitive data stores across cloud IAM, database roles, and SaaS permissions, identifies over-privileged access and dormant entitlements.

Assigns risk scores to discovered data based on sensitivity, access exposure, and configuration, then continuously monitors access patterns and policy compliance to surface the highest-risk data stores for action.

Acts on discovered data risks either natively or by orchestrating third-party DLP, IAM, EDRM, and ticketing controls: revoking over-permissioned access, quarantining or moving misplaced data, encrypting or masking unprotected files, and applying protection labels. Whether actions execute natively or only through integrated tools, and the breadth of available actions, are the primary differentiators; many DSPM products still leave enforcement to the integrated control.

Discovers and classifies sensitive data across a heterogeneous cloud estate in one inventory: object storage, managed data warehouses and lakes, cloud database services, and SaaS applications, including sources that are not supported out of the box through custom connectors. Breadth of supported sources and depth per source vary; on-premises and mainframe estates are covered under On-Premises and Mainframe Data Discovery.

Identifies sensitive data flowing into large language models and AI assistants such as Microsoft Copilot and ChatGPT, and enforces which generative AI services may use it, in which geographic region, and under which entitlements, reporting unsanctioned AI use. Right-sizing entitlements to stop oversharing before an AI assistant is rolled out is the most common form; blocking is usually delegated to DLP.

Baselines how users and service accounts normally access sensitive data stores and flags unusual access behavior in real time, such as mass downloads, off-hours access, or first-time access to regulated data, with detailed audit logs for investigating insider risk and compromised accounts. Often sold as data detection and response (DDR); products differ in whether detection uses ML baselining or static rules.

Identifies sensitive data in locations outside authorized data stores, development databases containing production PII, unprotected S3 prefixes, forgotten data lake partitions.

Traces the lineage of sensitive data across its life cycle, from origin through movements and transformations between storage locations, services, and users, surfacing unexpected cross-region transfers, shadow copies, and retention policy violations. Lineage depth (table and column level versus store level) varies; AI pipelines are covered under AI Pipeline Data Security.

Enriches classification results with context beyond the content itself, such as data lineage, effective permissions, storage location, owner, and business metadata, so that a record is labeled by what it is and how it is used rather than by pattern matches alone. Depth of contextual inputs, and whether they change the assigned sensitivity, vary widely across products.

Produces audit trails and regulation-mapped reports such as GDPR, HIPAA, and PCI DSS data inventories from discovery and access findings, with alerts on policy violations, so that evidence of data-handling practices can be handed to auditors without manual assembly. Custom and stakeholder-specific reporting is a common weak spot across products.

Discovers and classifies sensitive data held in on-premises estates without first migrating it to cloud: Windows file servers, SharePoint Server, NAS, self-managed relational databases such as SQL Server, Oracle, PostgreSQL, and MySQL, and mainframe environments including Db2. Cloud-first products often cover these sources slowly or not at all; depth of mainframe and legacy coverage is a primary differentiator.