This article summarizes key insights from an interview with Max Romanenko, Chief Engineering Officer at multi-model database provider EDB. The discussion focuses on how regulated enterprises can eliminate agentic AI risks including sensitive data leakage and non-compliant overreach actions through robust data and infrastructure-layer controls.
Core Differences Between AI Agents, Human Users, and Traditional Applications
AI agents differ fundamentally from human operators and conventional applications. Operating 24/7 with far higher execution speed and fidelity, AI agents are set to expand the global digital workforce by over 34% within three years — a scale of growth that took human workforce decades to achieve. This rapid adoption demands entirely new frameworks for access control, decision governance, and infrastructure design.
Traditional applications operate within fixed, predictable parameters, while human users act with clear role-based business context. In contrast, AI agents function without predefined boundaries: they autonomously interpret tasks, determine required contextual data, and retrieve live information without step-by-step human approval. Traditional credential validation is no longer sufficient. IT systems must additionally verify whether an agent’s behavior matches its assigned task scope. For example, a bank fraud detection agent should only access customer records relevant to its ongoing case, with strict in-line data governance blocking unrelated cross-data access.
Agent-Specific Identity and Access Management (Distinct from Human IAM)
AI agents require dedicated identity and contextual access control systems and cannot rely on conventional human IAM frameworks. Agents frequently execute hundreds of diverse tasks hourly, requiring dynamically adjusted, context-aware permissions centered on task-specific and scenario-based access precision.
A widespread enterprise security pitfall is granting agents broad service account privileges or full inherited human-user permissions, validating only credentials rather than the legitimacy of task-based data access. Agents with distinct business functions — such as bank fraud analysis and customer service — must be assigned isolated identities and tailored permission boundaries to build rigorous data security guardrails.
Defining and Enforcing Permission Boundaries for Cross-System AI Agents
AI agents operating across multiple systems require preemptive, purpose-bound access boundaries. Agents should only access data essential to their immediate task, even if the invoking user or application holds broader privileges. Permissions must be strictly limited to real-time task needs and expire upon task completion.
Generic application-layer policies fail to govern agent activity spanning APIs, retrieval systems, and databases. Security enforcement must shift to the data layer, where sensitive records reside and final access decisions are made. EDB leverages PostgreSQL’s native role-based permissions and row-level security (RLS) to enforce agent governance at the database level, rather than relying on aftermarket control planes. This approach balances security and performance, with a comprehensive agent governance layer scheduled for future release.
Mitigating Compliance Risks from Inherited Human Permissions
Legacy enterprise permission architectures are designed for trusted human employees and deterministic applications, not autonomous, dynamically behaving AI agents. This creates severe compliance risks for regulated industries: healthcare and financial service agents may legally access sensitive patient or customer data via valid credentials yet exceed permissible business scope, triggering compliance violations.
The definitive solution is to provision each AI agent with a unique independent identity, task-locked privileges, and separated read/write permissions. Database-level row-level controls enforce precise data visibility, preventing agents from becoming unregulated bypasses of enterprise compliance systems.
Audit, Row-Level Control, and Policy Standards for Regulated AI Environments
Compliance-grade agent auditing requires end-to-end traceability, fully logging agent identity, invocation source, accessed data, authorizing policies, and subsequent actions to reconstruct complete business context — a capability basic access logs cannot provide. Effective governance relies on native data-layer enforcement via PostgreSQL’s built-in controls, rather than retrospective auditing.
Systems must also differentiate risk levels between agent read and write actions with scenario-tailored permissions. For instance, customer data access for fraud investigation differs from routine customer service inquiries, just as clinical and administrative patient data access are isolated in healthcare. Row-level security blocks out-of-scope access at the source, restricting agents only to task-matched records and eliminating bulk sensitive data exposure risks.
Secure AI Adoption for Regulated Industries via Sovereign Data Infrastructure
Financial, telecom, and healthcare organizations must adhere to the core principle: bring AI to the data, not data to the AI. Regulated sensitive data must never be migrated to external, uncontrolled platforms, which breaks sovereign governance boundaries and introduces exposure risks.
Enterprises should deploy AI and analytics on top of their existing trusted, governed data foundations. Unifying transactional data, analytical workloads, and agentic AI on a single compliant architecture eliminates fragmented platforms and inconsistent control models, avoiding forced data migration and vendor lock-in.
Data Sovereignty: From Legal Terms to Technical Enforcement
Data sovereignty cannot remain merely a legal or contractual requirement — it must be embedded into runtime system logic. Legal agreements cannot restrict agent behavior, such as unauthorized data aggregation, cross-domain querying, or regulated data leakage. Regulated enterprises must enforce sovereign controls at the data layer via native database permissions and row-level security to deliver real-time, mandatory security constraints.
The Role of AI Agents in Sovereignty Governance
AI agents can assist data sovereignty management butmust never serve as the final authority. They effectively identify anomalous access patterns, flag misaligned data usage, and detect outdated configurations, reducing the burden of manual inspection for complex modern data environments.
However, core sovereignty rules and security boundaries must be defined and controlled by human administrators. AI agents operate strictly within enterprise-set governance frameworks and have no authority to override or bypass established compliance and sovereignty policies.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Vietnamese
