Webinar Description
Key Takeaways
- Explores runtime data access control for AI agents operating in production environments
- Addresses security risks posed by overprivileged service accounts across cloud data platforms
- Covers identity-aware access enforcement from data products down to individual fields
- Relevant to organisations using Snowflake, Databricks, and data lake architectures
- Designed for CISOs, data architects, security engineers, and compliance professionals
Introduction
As AI agents move from experimental deployments into production workloads, organisations face a growing security challenge: these agents typically operate with broad service account privileges that grant access to sensitive data across multiple platforms. A webinar hosted by DataHub and SecuPi examines how enterprises can extend governance frameworks into runtime environments, ensuring AI agents access only the data they are authorised to use based on end-user identity and contextual authorisation policies.
The session targets data security and governance professionals grappling with the operational reality that traditional access control models were not designed for autonomous AI workloads. With regulatory scrutiny intensifying around automated decision-making and data protection requirements remaining stringent, the question of what AI agents can access—and under what circumstances—has become a pressing concern for enterprise security teams.
About This Event
This virtual webinar brings together perspectives from DataHub and SecuPi to demonstrate approaches for enforcing granular access control over AI agents at runtime. The session focuses on practical implementation of need-to-know access policies that operate at multiple levels of granularity, from broad data products down to individual data fields.
Rather than relying solely on static permissions assigned at deployment, the approaches discussed emphasise deterministic runtime control—access decisions made dynamically based on the specific context of each data request, including the identity of the end user on whose behalf an AI agent operates.
The Challenge of Overprivileged AI Agents
AI agents entering production environments typically inherit service accounts with extensive permissions across data infrastructure. These accounts may have legitimate access to Snowflake warehouses, Databricks workspaces, data lakes, and various enterprise applications—far exceeding what any individual query or task requires.
This architectural pattern creates significant risk. An AI agent processing a customer service request, for example, might technically have access to financial records, employee data, or strategic business information entirely unrelated to its immediate task. The principle of least privilege, long established in security practice, becomes difficult to enforce when agents operate autonomously across interconnected data platforms.
The webinar addresses how organisations can implement access controls that respect authorisation boundaries without requiring fundamental changes to existing data infrastructure or creating operational bottlenecks that undermine the efficiency gains AI agents are meant to deliver.
Runtime Governance and Context-Aware Access
Central to the discussion is the concept of extending governance beyond static policy definitions into active runtime enforcement. Traditional data governance frameworks excel at cataloguing data assets, defining ownership, and establishing access policies—but enforcement often occurs only at connection time or through periodic access reviews.
Runtime data access control introduces continuous policy enforcement, evaluating each data request against current authorisation context. For AI agents, this means access decisions can incorporate the identity of the human user initiating a workflow, the specific data elements required for the task, and organisational policies governing sensitive information categories.
This approach enables what the session describes as deterministic runtime control—predictable, auditable access decisions that security teams can verify and compliance officers can document.
Who Should Attend
The webinar is designed for professionals responsible for securing AI deployments and managing data access across cloud platforms. Chief Information Security Officers and Chief Technology Officers evaluating governance frameworks for AI initiatives will find relevant strategic context. Data platform architects and security engineers seeking implementation approaches for runtime access control represent the primary technical audience. Compliance officers concerned with demonstrating appropriate data protection controls for automated systems will also benefit from the discussion of auditable, policy-driven access enforcement.
Conclusion
As AI agents assume greater operational responsibilities within enterprise data environments, the gap between their technical access capabilities and appropriate authorisation boundaries presents measurable risk. This webinar offers practitioners an opportunity to examine how runtime data access control can address that gap, maintaining the productivity benefits of AI automation while preserving the governance controls that regulatory requirements and organisational policies demand.