Oracle is putting AI agents inside a tightly controlled security workflow. On October 27, 2026, at Oracle AI World in Las Vegas, the company announced Oracle AI Security Mesh, a managed platform powered by OpenAI’s frontier cyber models and designed to help organizations find and fix security vulnerabilities faster.
The platform brings testing, evidence, and agent-based workflows into one system, while keeping customers in control of approved scopes, permissions, and sensitive actions. That combination creates a clear question for enterprise security teams: can AI investigate complex weaknesses without gaining permission to change systems on its own?
A Security Platform Built Around Controlled AI Agents
Oracle AI Security Mesh covers applications and infrastructure running on-premises, in OCI, or in other clouds. The platform pairs OpenAI’s frontier cyber models with Oracle’s testing tools and the context customers provide about their own systems.
Oracle said the models reason through application behavior and investigate how potential weaknesses could be exploited. The platform then helps security teams prioritize the most critical risks, giving teams a path from discovery to evidence and proposed fixes instead of leaving them with a long list of unverified alerts.
Each finding includes supporting evidence, such as reproduction steps and test results. Teams can use that evidence to evaluate proposed fixes, while the platform checks whether a fix affects expected application behavior.
Testing happens inside a scope the customer approves. No remediation change is applied until the customer approves it, creating a human decision point before any change reaches the system.
Every agent also runs under a distinct identity with a defined task scope and just-in-time permissions. Independent monitors can detect and stop activity outside approved boundaries, adding another control layer around agent activity.
From System Discovery to Confirmed Weaknesses
Oracle’s product page describes discovery frameworks that map a customer’s systems and the connections between them using the context and access the customer provides. Those frameworks help testing agents find weaknesses with less manual setup, allowing the platform to build its investigation around the customer’s actual environment.
Specialized agents trace how weaknesses connect across systems and test possible attack paths. A separate agent checks whether findings can be reproduced, helping distinguish a confirmed problem from a weakness that only looks serious during an initial review.
Discoveries remain in a private knowledge base for each business. The platform prioritizes the most serious confirmed weaknesses, then gives teams a chance to review recommended fixes and agent-assisted code changes before those changes are retested.
- Discovery frameworks map systems and connections from customer-provided context and access.
- Testing agents investigate weaknesses and possible attack paths.
- A separate agent checks whether findings can be reproduced.
- Evidence includes reproduction steps and test results.
- Teams approve remediation changes and review agent-assisted code changes before retesting.
Jordan Acosta, a product marketing manager for AI at Oracle, described the workflow in a blog post published October 27, 2026. Oracle maps a customer’s systems and proposes a test plan for approval, after which specialized AI agents investigate weaknesses and test how someone could exploit them.
A separate agent repeats the test to confirm the problem is real. Coding agents can then help prepare code fixes for the team’s approval, keeping the security team involved before changes move forward.
Security Controls Designed for AI Workloads
Oracle AI Security Mesh complements OCI Identity Assurance, which uses biometrics to verify human identities. It also complements OCI Zero Trust Packet Routing, which enforces network access policies to protect critical systems and data.
Together, these capabilities address different parts of the same challenge: verifying human identities, controlling network access, and managing what AI agents can do inside business systems. Phil Newman, Oracle’s senior vice president of Security and Identity, said, “As AI agents take on more responsibility, enterprises need security that can keep pace with how these systems operate,”
Newman said the combined capabilities are intended to help organizations identify and resolve security risks faster while maintaining control over who can authorize sensitive actions and which systems AI workloads can access.
That focus on authorization runs through the entire platform. Customers approve the testing scope, agents receive defined permissions, monitors can stop activity outside approved boundaries, and remediation waits for customer approval.
Oracle’s announcement places AI security inside a managed process rather than treating cyber models as standalone tools. The platform is designed to connect system discovery, weakness testing, evidence collection, fix evaluation, and retesting in a sequence that gives security teams control at each important step.
As organizations use AI agents across applications and infrastructure, that workflow points toward a new model for enterprise defense: agents investigate and prepare the work, while customers decide what the agents may access and which changes may proceed.
Based on
