SecuPi monitors the end-user’s identity behind MCP and NHI database access, extending DAM with runtime AI Data Access controls

SecuPi today announced new capabilities designed to close a rapidly growing security and compliance gap in traditional Database Activity Monitoring (DAM): AI agents accessing enterprise databases through Model Context Protocol (MCP) servers using shared non-human identities (NHI) and anonymous service accounts.

As AI agents proliferate, MCP servers become a new path into enterprise databases. These servers usually connect through shared service accounts, causing traditional DAM tools and native database audit logs to record the anonymous service account/NHI while losing visibility of the end user context, AI agent and purpose behind the database activity.

The New DAM blind spot: AI Agents behind shared NHIs

Traditional DAM was designed primarily around identifiable human users, applications and database admin tools.

AI agents introduce a different access model:

End User → AI Agent → MCP Server → NHI / Service Account → Database

When hundreds or thousands of users and AI agents access databases through shared service accounts, DAM agents and native database logs may see the account executing a query — but not the human or agent responsible for the request.

This creates a fundamental audit gap for security and compliance teams.

SecuPi restores the missing identity context

SecuPi provides an MCP-aware audit service that creates a tamper-resistant audit trail linking the end user to the AI agent, MCP server, NHI/service account, the sensitive data accessed and database action performed.

Instead of seeing only:

Service Account → Database Query

security teams gain real-time visibility into:

End User → AI Agent → MCP → Service Account → Sensitive Data → Action

This enables organizations to overcome the identity limitations of native database logs and extend DAM with the user and AI agent context required for effective AI database auditing.

Detect and prevent anonymous or high-risk AI Agent activity

SecuPi continuously identifies and classifies AI access to sensitive data and assigns risk based on user identity, agent identity, data sensitivity, actions and behaviors.

SecuPi policies can dynamically:

• Alert on unidentified or suspicious AI agent activity
• Quarantine untrusted agent access
• Block anonymous AI access to sensitive databases
• Identify sensitive data accessed by AI agents
• Classify database actions and assign risk scores
• Enforce object-, column-, row- and field-level access controls
• Apply dynamic masking, filtering, format-preserving encryption and tokenization
• Maintain a tamper-resistant AI and database audit trail for forensics and compliance reports

From Database Activity Monitoring to runtime AI control

For highly sensitive databases, monitoring alone is no longer sufficient.

SecuPi extends traditional DAM from the observation of database activity to preventive runtime data security, directly enforcing policies on what AI agents are allowed to access, see and do.

Based on the end user, agent, purpose and data sensitivity, SecuPi can dynamically:

• Mask sensitive data
• Filter rows or fields
• Tokenize or encrypt data
• Restrict actions
• Block access

These controls enable enterprises to enforce ‘least privilege’ and ‘need to know’ even when many AI agents operate through the same database service account or NHI.

DAM must know the human behind the AI agent

AI agents represent a new database access layer that traditional DAM architectures were not designed to understand.

Database Activity Monitoring must go beyond recording database queries and establish three critical dimensions of AI accountability:

• WHO: the human and AI agent behind the NHI
• WHAT: the sensitive data the agent accessed
• DO: the actions the AI agent performed

SecuPi brings these dimensions together through AI identity-aware auditing, sensitive-data intelligence and deterministic runtime access controls.

The result is a DAM architecture that helps organizations safely expose enterprise databases to agentic AI while maintaining end-user accountability, sensitive-data protection and complete AI agent auditability.