Trust3 AI has announced AgentDOS, an enterprise control plane that provides visibility into AI agents, including real-time token consumption monitoring across platforms such as Databricks Agent Bricks and Microsoft Copilot Studio.

As enterprises rapidly scale AI adoption, a new class of risk is emerging: autonomous agents acting beyond their intended scope, accessing regulated data, and silently driving up token consumption. Regulatory frameworks such as the EU AI Act and standards like the NIST AI Risk Management Framework are raising the stakes further, requiring organizations to demonstrate accountability and control over automated decision-making systems.

Observability tools were built for developers debugging models. They were not built for security, governance, and operations teams responsible for compliance, cost control, and enterprise risk.

AgentDOS addresses this gap by giving enterprises a unified view of every AI agent, every action, and every token consumed. Built as part of Trust3 AI’s One Control Plane architecture, AgentDOS extends the company’s Unified Trust Layer to bring governance, security, and observability together across any agent framework, cloud, or data environment. Databricks Zerobus or Apache Kafka processes high-volume logs and traces, including native OTel telemetry, and aggregates signals across platforms into a single unified console.

“Enterprises don’t have an AI problem. They have an AI visibility problem. Agents are already making decisions, accessing sensitive data, and consuming budgets without oversight. AgentDOS gives security and governance teams the control plane they’ve been missing,” said Balaji Ganesan, CEO, Trust3 AI.

Token consumption is quickly becoming one of the largest and least visible cost centers in enterprise AI. AgentDOS introduces policy-driven token observability, allowing organizations to track, control, and enforce usage limits across agents in real time, preventing budget overruns before they occur.

With AgentDOS, enterprises can:

• Detect scope drift in real time, identifying agents operating outside approved boundaries before risk escalates
• Monitor token consumption across platforms, enforcing policy limits across Databricks Agent Bricks, Microsoft Copilot Studio, and more
• Automatically discover and inventory agents, assigning dynamic trust scores across security, compliance, and accountability dimensions
• Trace every agent decision with full fidelity, including prompts, retrievals, tool usage, and data access, all replayable on demand
• Gain visibility into regulated data access, understanding exactly which datasets agents are interacting with in production

Unlike legacy tools, AgentDOS is designed for security, compliance, and governance teams, not just developers. Every agent action is enriched with identity, declared purpose, data lineage, and live policy context, enabling organizations to enforce governance in real time rather than after the fact.

A healthcare provider managing sensitive patient data across multiple systems deployed AgentDOS to govern a growing fleet of AI agents running on Azure and Databricks. Within days, the organization identified multiple agents operating outside their declared scope, including two accessing regulated patient datasets without valid purpose context.

At the same time, token consumption monitoring flagged an agent on track to exhaust its monthly allocation in just 11 days, an overrun that would have gone undetected with existing tools. By shifting from manual audits to real-time observability powered by OTel telemetry, the organization significantly reduced HIPAA audit preparation time, eliminated unplanned AI spend, and gave the CISO a unified view of all agent activity touching protected health information.