Multi-Tenant AI Agent Data Isolation | CockroachDB
Blog post from Cockroach Labs
Multi-tenant AI agents can magnify data-access failures because improperly retrieved information may enter an agent’s reasoning process and trigger downstream tool calls, API actions, or writes at machine speed, unlike traditional SaaS errors that are generally limited to a single response. Citing recent prompt-injection, privilege-escalation, and supply-chain incidents involving Salesforce Agentforce, ServiceNow, and Salesloft/Drift, the discussion argues that application-layer tenant filters alone are unreliable because developers, ORMs, dynamically generated queries, or attacker-controlled tool parameters can bypass or omit them. It presents database-enforced row-level security (RLS), particularly in CockroachDB shared-schema deployments, as a way to apply tenant-specific policies automatically to every query, provided applications set tenant context correctly and agent roles lack RLS-bypass privileges. The article also describes CockroachDB’s REGIONAL BY ROW placement as a means to keep each tenant’s data in a designated geographic region for residency requirements without separate regional clusters. It emphasizes that RLS is only one layer of agent security and does not prevent prompt injection within a tenant’s data, shared inference-cache or vector-store leaks, overly broad service accounts, or possible bypasses in specific database access paths, making least privilege, cache partitioning, tool controls, and validation necessary complementary protections.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.