Partnership announcement

Operant and Databricks partner to secure AI agents.

Runtime defense for everything running on Unity AI Gateway.

Joint customers can now connect the Operant Semantic Firewall directly to Databricks Unity AI Gateway. Every model request, model response, MCP tool call, and tool result can be evaluated by Operant before it reaches its destination, without changing a line of application code.

Today, we at Operant are proud to announce a partnership with Databricks. Organizations that use both Operant and Databricks can now connect the Operant Semantic Firewall directly to Databricks Unity AI Gateway, where it works in step with the gateway on every request. Once connected, every model request, model response, MCP tool call, and tool result flowing through the gateway can be evaluated by Operant before it reaches its destination or returns to the caller.

For joint customers, that means one thing: the AI applications and agents you build on Databricks can be defended at runtime by Operant, without changing a line of application code.

Why this matters now

AI applications are quickly becoming agents: systems that don't just answer questions but query databases, call internal tools through MCP, and act on what they find.

That shift changes the security problem. The risk is no longer limited to what a model says, or even to what's inside a model file. It's what an agent does: a prompt that tricks it into exporting customer data, a poisoned tool argument that triggers a destructive operation, a tool result that smuggles malicious instructions back into the model. These attacks arrive through the same traffic as legitimate work, and they move at machine speed.

Unity AI Gateway provides a control point for AI traffic on Databricks. Operant, connected to the gateway, adds the intelligence to understand what each request and response is actually trying to do, and to stop the ones that shouldn't happen.

“Agents don't fail safely. When one is manipulated, it doesn't just say the wrong thing; it calls the wrong tool, touches the wrong data, and acts at machine speed. Operant was built to stop that at runtime, by understanding the intent behind every request and every tool call. Bringing that to Databricks Unity AI Gateway means joint customers can secure their agents where they already route AI traffic, without rewriting a single application.”

— Vrajesh Bhavsar, CEO and Co-founder, Operant AI

Better together

The partnership pairs two independent platforms, each doing what it does best.

Databricks remains the enforcement point. Unity AI Gateway continues to route traffic, apply policies, and enforce outcomes for every governed service. Customers keep their existing Databricks workflows, access controls, and operational model.

Operant provides the security decision. When the gateway sends a request or response for evaluation, the Operant Semantic Firewall analyzes it against the organization's active Operant security policies and returns a verdict: allow or deny, with a human-readable reason Databricks can pass back to the caller.

The result is runtime AI defense from Operant, applied through the gateway Databricks customers use for AI traffic. Developers don't route traffic somewhere new or call another API. Platform teams connect Operant once in Unity AI Gateway, and every governed service they attach it to is covered.

This is available today, not a roadmap. Operant evaluates traffic inline and returns enforceable verdicts on every model request, model response, MCP tool call, and tool result, so threats are stopped before they reach a model, a tool, or a user, not just observed after the fact.

“Customers are putting agents to work on their most important data and systems, and they want to do it securely from day one. Operant brings the intelligence to understand what every model request and MCP tool call is trying to do, and to stop what shouldn't happen before it reaches a model or a tool. Connected to Unity AI Gateway, that protection goes wherever our joint customers' AI traffic goes, giving them the confidence to move agents into production and keep them secure as they scale.”
— Stephen Orban, SVP, Product Partnerships and Ecosystem, Databricks

What Operant protects

The integration covers the full path of an AI interaction, in both directions, for both models and tools.

Model requests
Operant scans system instructions, conversation history, and user prompts, including text inside structured and multimodal message blocks, across common provider request formats. It detects prompt injection, sensitive-data exposure, unsafe content, and malicious code or commands.
Before the model runs
Model responses
Generated output is evaluated for sensitive information, secrets, unsafe content, and policy violations, using the original request as context so the verdict reflects the full exchange.
Before they return
MCP tool calls
Operant evaluates the MCP service, the tool's identity, and its complete arguments. Policies on a restricted or destructive tool still apply when it's called with no arguments at all, closing a gap that argument-only inspection leaves open.
Before the tool runs
MCP tool results
The complete result is scanned, including structured fields, so malicious or sensitive content can't hide alongside otherwise benign text. Tool results are one of the most direct paths for injecting instructions into a model mid-task.
Before it reaches the model or caller

Ready for production from day one

Start in log mode, then enforce

Operant evaluates live traffic and records its real verdicts without blocking anything. When you're ready, enforce mode applies the same verdicts.

Detection and enforcement are separate

Record a finding without blocking traffic when no enforcement policy requires it. Teams decide which risks to observe and which to stop.

Policies fit each service

Each policy attaches to a specific governed service, with input and output evaluation enabled independently.

Fail-closed by design

If an evaluation times out, receives a malformed event, or hits an internal error, the result is deny. Gaps in evaluation never become gaps in protection.

Secure, simple connection

Databricks and Operant connect server-to-server with OAuth machine-to-machine authentication, configured in Unity AI Gateway as an external guardrail.

Available today, not a roadmap

Inline, enforceable verdicts stop threats before they reach a model, a tool, or a user, not just observed after the fact.

Why Operant

Runtime defense for every model. Operant secures AI at runtime across thousands of models, wherever they run. Through Unity AI Gateway, that protection extends to every model and MCP service a Databricks customer attaches it to.

Security for what AI does, not just what it's made of. Scanning a model before deployment checks the artifact, but it can't see what happens once that model is running inside an agent: the tools it calls, the data it touches, the code it executes, the scope it operates in, and the intent behind each action. That's where today's attacks happen, and that's where Operant works. Operant's AI supply chain coverage follows the same principle, securing the models, tools, and MCP servers agents depend on as they're actually used.

Intent, not just patterns. The Operant Semantic Firewall, built on patented technology proven against rogue agents, evaluates what each request and response is trying to accomplish, rather than matching it against known attack patterns.

Deep MCP coverage. Operant protects the MCP attack surface in depth, including tool calls with no arguments and threats hidden in structured tool results.

Visibility built for audits and the SOC. The Operant 3D Runtime Defense Platform gives security teams live visibility into AI activity across their environment, with records that support audit and compliance programs. Findings flow into the SIEM and SOC tools security teams already use, so AI threats are investigated through established processes.

What joint customers get

  • Ship agents faster, with security already in place. Developers keep building the way they do today. Operant's protection is applied through the gateway, not through changes to every application.
  • One control point for models and tools. Connected to Unity AI Gateway, Operant covers model services, model provider services, and MCP services, so agent tool use is protected alongside model traffic.
  • Defense that understands intent. The Operant Semantic Firewall catches attacks that pattern matching misses, such as an innocuous-looking request to summarize a sales database and send it to an outside domain.
  • A rollout security teams can trust. Log mode, per-service policies, and fail-closed evaluation let teams move from visibility to enforcement on their own schedule.
  • Evidence and response, not just blocking. Runtime visibility supports audits, and findings reach the SOC through existing SIEM workflows.

Get started

The integration is available now to organizations with both an Operant subscription and a Databricks workspace with Unity AI Gateway enabled. Operant customers connect Databricks from the Integrations page in the Operant console, generate partner guardrail credentials, and attach Operant as an external guardrail to any governed service in Unity AI Gateway. The setup guide is available in the Operant documentation.

New to Operant? Request a demo to see the Operant Semantic Firewall protecting Databricks AI traffic.

Secure your agents on Databricks.
Without rewriting a single app.

See the Operant Semantic Firewall protecting Databricks AI traffic.