VAST Data debuts AI tools for safer, smarter agents

VAST Data has announced two new services for its VAST AI Operating System aimed at governing AI agent activity and improving model performance through continuous tuning.

The products, PolicyEngine and TuningEngine, sit within the VAST DataEngine. VAST Data positions them as building blocks for organisations seeking larger AI deployments with tighter oversight and more predictable behaviour.

AI governance is a growing concern for business leaders across the Asia-Pacific region. VAST Data cites research showing 70% of organisations view AI risk and opportunity as the most pressing board topic for 2026. The focus, it argues, is shifting from building models to ensuring AI workloads follow rules, decisions can be traced, and performance improves over time.

Policy controls

PolicyEngine targets the risks created when AI agents and workflows interact with corporate data and other systems. Agents can generate new data such as responses, agent-to-agent messages, and event logs, increasing the number of places where sensitive information can appear.

Without granular controls over what agents can access and how they communicate with tools, remote data products, and other agents, the risk of data leakage rises. VAST Data also says stronger workflow logging is needed for systems that can be audited and explained.

PolicyEngine is designed as an inline policy enforcement layer that checks access, actions, and communications against explicit permissions and context before an action executes.

It also maintains traces and logs that VAST Data describes as tamper-proof, supporting what it calls a "zero-trust operating posture" for its AI environment. This aligns with how many security teams now treat internal services, with verification and logging applied as standard controls.

Continuous tuning

TuningEngine addresses a different operational challenge: many organisations deploy a model and then improve it through separate cycles of data collection, training, and validation. VAST Data is proposing a tighter loop that captures outcomes from AI agent workflows and feeds them directly into model improvement.

Within this approach, VAST Data's AgentEngine provides the agent runtime for the AI operating system. It is described as a serverless environment that coordinates multi-agent workflows, model invocation, and tool usage. The platform has been used to deploy static models and is now being extended to support learning loops based on telemetry and feedback.

TuningEngine collects outcomes from agent pipelines and applies curated feedback to improve model performance over time. VAST Data says it supports methods including LoRA fine-tuning, supervised fine-tuning, and reinforcement learning.

In the workflow described, tuning pipelines ingest data, process it, and propose candidate models. These candidates can be evaluated and benchmarked within the VAST AI Operating System before deployment. Deployment can be manual or automatic, after which new interactions feed the next improvement cycle.

Platform strategy

VAST Data's broader message is that operational AI needs stronger guardrails and a more integrated feedback model. It frames this as a move toward systems that "observe, reason, act, evaluate, and improve" while maintaining security and explainability across workflows in a single environment.

In that strategy, PolicyEngine and TuningEngine are intended to work together: one defines and enforces acceptable behaviour and access, while the other iterates on model quality using performance signals and real-world feedback.

Jeff Denworth, co-founder at VAST Data, said the announcement reflects a wider shift in how organisations treat AI applications in production.

"Just as people are always learning, so should tomorrow's applications," Denworth said. "With the introduction of PolicyEngine and TuningEngine, the VAST AI Operating System has become a thinking machine that customers can deploy wherever they compute - a machine that safeguards every interaction and learns from every outcome, bringing the power of AI within reach of every organisation."

VAST Data announced the services at its Forward conference in the US. PolicyEngine and TuningEngine are slated for release by the end of 2026.