Oracle’s Secure at Source approach puts encryption and access controls inside databases, shifting enforcement away from individual applications, SiliconANGLE reported.
Key points
- Oracle says database-level controls can apply the same permissions across applications and agent access paths.
- In an Oracle demonstration, a prompt exposed salary data through an application but failed when the database enforced access restrictions.
- Oracle Deep Data Security is designed to restrict access at the row, column and cell level according to user identity and context.
Secure at Source moves enforcement into Oracle databases
Vipin Samar, Oracle’s senior vice president of Database Security, said Secure at Source uses the database as a common point for enforcing permissions. The approach is intended to keep restrictions attached to users and data when access comes through different applications or AI agents. It also places protections other than encryption inside the database, he said.
At Oracle’s “AI Cyberattacks Are Escalating: How to Secure Your Data Now” event, Samar argued that agents should not be responsible for enforcing their own access limits. Oracle’s approach, he said, is to restrict an agent to information the user is authorised to obtain, regardless of the route the agent takes to the database.
An Oracle demonstration put those two enforcement points to the test. An indirect prompt got past application guardrails and retrieved salary data. Repeating the prompt with restrictions applied inside the database prevented access to the protected information, SiliconANGLE reported. The test was an Oracle demonstration, rather than a reported customer deployment.
David Knox, Oracle’s vice president of Database Security product management, described enforcement at the point where the database handles a query. Oracle can intercept a SQL query as it executes and apply its security capabilities there, he said. That places the access decision beneath the application presenting the result to a user or agent.
Deep Data Security sets row-level permissions
Juan Loaiza, executive vice president of Oracle Database Technologies, described a related control in Oracle Deep Data Security. The product enforces authorisation at the row, column and cell level using a user’s identity and the context in which a request is made. Those restrictions are intended to limit what an agent can retrieve if a prompt is manipulated or an application is compromised.
Application code has traditionally determined which records a person can see and which transactions they can carry out, Loaiza said. Oracle’s proposed arrangement moves that decision closer to the stored information. Under the approach he described, an agent acting for an end user would be able to see only data that user is permitted to access.
Loaiza said putting authorisation in the database could reduce the security work assigned to individual application developers. His concern was that a user could phrase a prompt in a way that causes an AI system to take an action its application-level instructions were meant to prevent. Database permissions would still apply when the agent attempted to retrieve the data, he said.
The distinction extends beyond a single agent interface. Samar said AI can move between databases, making consistent controls across a customer’s estate important to Oracle’s approach. Applying restrictions at access time addresses the query itself, while fleet-wide tools give administrators a way to inspect users, policies and database configurations.
Data Safe and Security Central cover database fleets
Oracle Data Safe provides cloud-based security assessment and monitoring, while Oracle Database Security Central offers a self-managed view across databases, according to Samar. He said Security Central brings together information about users, sensitive data, configurations and policies. Oracle positions those tools for organisations operating databases across cloud and on-premises environments.
Fleet oversight is meant to help administrators find changes in configuration, excessive access privileges and differences between policies. Knox described that centralised management alongside controls applied when a database is accessed. The products therefore address separate parts of Oracle’s approach: reviewing the condition of a database fleet and checking access when a query runs.
Loaiza also said Oracle is using multiple AI models to examine its own code for security vulnerabilities and is urging customers to keep database environments current. He said AI had found hundreds of security issues in Oracle’s products. His account placed software updates alongside access controls as another part of the company’s database security work.
SiliconANGLE reported that Oracle’s Zero Data Loss Recovery Appliance protects database transactions in real time, checks backup integrity and supports rapid restoration. He also said organisations should isolate trusted backups and plan to rebuild operations without restoring malicious code or hidden access points.
