The Pulse
Rocket Puts Policy Gates Between AI Reasoning and Mainframe Execution
Rocket Software expands its EVA agentic AI platform for mainframe operations and introduces PlanGuard, a policy layer between AI decisions and system execution. The company says global pilots are testing governed automation across financial
AI.info Team ·
Rocket draws a line before execution
Rocket Software is expanding its Rocket EVA agentic AI platform with a control layer designed to stop autonomous reasoning from turning directly into unapproved action on a mainframe.
The company announced Rocket PlanGuard on September 23, describing it as a security layer that places a policy checkpoint between an AI agent’s reasoning and system execution. PlanGuard adds policy decisions, identity controls and activity logging so organizations can restrict what an agent may do, when it may do it and which mainframe resources it can reach.
Rocket’s pitch addresses a central tension in applying AI to core systems: the more useful an agent becomes, the more access it needs. “Enterprises have relied on the mainframe to run their mission-critical workloads for decades,” Milan Shetti, president and CEO of Rocket Software, said in the announcement.
PlanGuard makes approval part of the agent workflow
Rocket says EVA can correlate operational data, reason across system context and automate or recommend tasks within boundaries defined by an enterprise. PlanGuard is intended to enforce those boundaries through contextual, just-in-time authorization rather than relying only on a broad set of standing permissions.
The company also says actions are logged for audit purposes, with human oversight available where required. That combination gives security and operations teams a record of what an agent attempted or completed, while allowing them to define the scope of its access.
Rocket does not describe PlanGuard as a separate standalone product in the announcement. Instead, it presents the technology as a security layer within EVA, which already connects conversational AI with mainframe operational data through a model-agnostic, standards-based architecture.
EVA moves from diagnosis toward action
Rocket launched EVA in January as a conversational assistant for operational diagnostics. The expanded platform broadens that role to include agents that can reason over operational context and take or recommend governed actions.
The use cases named by Rocket range from job-failure pattern analysis and operational dashboards to end-of-month financial reporting, CICS application optimization and queue analysis. Other examples include vulnerability detection, compliance automation, patch management, batch-performance monitoring, high-CPU analysis and IBM Db2 data-sharing diagnostics.
Those examples place EVA close to the daily work of mainframe operations rather than treating the system as a general-purpose chatbot. Rocket says the platform can work with z/OS data sources including SMF records, job output and message queues, giving agents access to the operational signals needed for investigation and response.
A skills shortage supplies the business case
Rocket is tying the product expansion to a mainframe staffing problem. The company cites Hanover Research, which found that 81% of financial-services IT leaders describe their mainframe skills gap as very or extremely significant, while 87% believe AI will help address it over the next two years.
The same research says 94% of financial-services IT leaders rank improving IT operations with AI as a high or top priority. Those figures come from research commissioned by Rocket, so they describe the market through the company’s chosen sample and framing rather than an independent industry census.
EVA is designed to preserve institutional knowledge by making operational expertise available through natural-language interaction. Rocket says that can help less-experienced staff investigate incidents while reducing reliance on a shrinking pool of specialists, without changing production workloads.
Global pilots will test the boundaries
Organizations in financial services, government, insurance, retail and telecommunications are participating in Rocket EVA pilots, according to the announcement. Rocket says the focused program lets customers use their own data to validate operational use cases and move from installation to actionable insights within days rather than weeks.
The pilots will determine whether the policy model works under the conditions that make mainframes difficult targets for automation: tightly controlled access, long-running workloads, multiple dependencies and audit requirements. Rocket’s announcement does not provide customer names, deployment volumes, pricing or a general availability date for the expanded capabilities.
For now, the concrete change is PlanGuard’s position in the execution path. EVA may reason about a mainframe environment, but Rocket says the agent must pass through an enterprise-defined policy checkpoint before it can act.