Netzilo brings AIDR runtime governance to Amazon Bedrock AgentCore

Netzilo has extended its AIDR runtime governance and enforcement layer to Amazon Bedrock AgentCore, Microsoft Foundry, Microsoft Copilot Studio, CrewAI, LangGraph, and Google Vertex AI, making Bring Your Own Governance the default for any agent in a supported harness. The expansion lands as enterprises move agents from pilots into production, and security teams ask for a single, portable way to observe, detect, and respond across every framework an agent might run in.

"AI agent governance cannot depend on which platform exposes which integration point," Egemen Tas, CEO of Netzilo, told Help Net Security. "Enterprises need governance that follows the agent wherever it runs. With Netzilo AIDR, organizations can bring their own governance to Amazon Bedrock AgentCore and the broader agent ecosystem without accepting fragmented visibility, degraded enforcement, or platform-specific blind spots."

The runtime behavior graph behind Netzilo AIDR

Netzilo AIDR builds a runtime graph of agent behavior, including tool calls, file reads, network requests, skill acquisitions, and multi-stage action sequences. The platform correlates behaviors that look harmless in isolation but turn into risk when read together, helping security teams detect prompt injection, indirect prompt injection, tool poisoning, capability hijacking, privilege escalation, and multi-stage data exfiltration in real time. The runtime graph is the load-bearing technical claim of the announcement, because every other guarantee in the product follows from being able to see the full sequence of actions an agent took before a sensitive call, and not just the call itself.

Rather than relying on the telemetry, policy hooks, or native controls exposed by each agent platform, Netzilo brings its full behavior graph and enforcement layer to the agent itself. Every agent receives the same level of behavioral visibility, detection, correlation, and response regardless of the harness where it runs, which removes the platform-specific blind spots that have made agent visibility a patchwork to date. For security teams that already write policy in code, the move from per-platform integration to portable runtime enforcement is the load-bearing change in this announcement, and it is the part that turns AIDR from a logging product into a control plane. A runtime graph that lives inside the agent itself, rather than being reconstructed from the platform's own logs, is also the only way to correlate tool calls, file reads, and network requests across the multi-stage sequences that real agent work produces.

The practical effect for a security team is that a single Network of policy primitives, expressed in the same Governance-as-Code dialect, can be applied to a coding agent in Amazon Bedrock AgentCore, a customer-facing assistant in Microsoft Foundry, and a back-office data agent in Google Vertex AI without rewriting the underlying detection rules. That portability is what the integrations list is built to demonstrate, and it is the architectural choice that most distinguishes AIDR from the per-platform agent governance tools that have shipped earlier in 2026.

Governance-as-Code and the agentic control plane

AIDR also enforces deterministic Governance-as-Code controls and can isolate or terminate a compromised agent in real time. That positions the platform as a runtime control plane for the agentic workforce, extending beyond prompt-level supervision or static access control. A governance layer that cannot stop an agent is, in practice, a logging layer with a different name, and the isolate-or-terminate capability is the specific line the announcement draws between runtime governance and the prompt-only supervision that the agent platform vendors ship by default. The same enforcement primitives can be applied to a coding agent, a customer-facing assistant, and a back-office data agent without rewriting the underlying policy, which is the portability claim that the integrations list is built to demonstrate.

The same architecture extends Zero Trust to autonomous AI agents without forcing the enterprise to surrender data, policy, or operational control to any single platform. Agents running on user devices, mobile phones, and on-premises harnesses can sit under the same governance surface as agents in Amazon Bedrock AgentCore or Google Vertex AI, which is the deployment topology most enterprise security teams have been asking vendors for since the first wave of agent platform announcements in early 2026. Netzilo's framing of the move is that the agent, not the harness, is the unit of governance, and that posture is what allows the platform to cover Bedrock, Foundry, Vertex, CrewAI, and LangGraph without producing a per-vendor product line. It also keeps the enterprise's data and policy under its own control, which is the regulatory and procurement argument that the announcement leans on hardest in the European and APAC markets where data-sovereignty rules are the tightest.

The runtime control plane also gives security teams a uniform place to revoke an agent's tool access in response to an incident, audit which agent ran which sequence of actions, and replay an agent's behavior in a forensic investigation. Each of those primitives is a real product surface for AIDR, and each is the kind of capability that platform-level observability tools typically do not expose. The agentic workforce, in Netzilo's framing, is being treated the same way that the human workforce has been treated for the last twenty years of identity and access management, and the Governance-as-Code primitives are the agentic equivalent of the IAM policies that security teams have been writing since the early 2000s.

Where Netzilo AIDR fits in the 2026 agent governance stack

Netzilo is not the first vendor to ship runtime governance for AI agents. Snyk shipped Evo ADS as a governance layer for AI coding agents on June 23, the MCP supply chain had 71 percent of public packages with a single maintainer as of mid-June, and Microsoft Agent 365 went GA earlier in May as a control plane for the Microsoft agent stack. What AIDR is adding in the same week is portable runtime enforcement across the agent harnesses enterprise teams are actually using, rather than a governance layer tied to one coding tool or one runtime. The integrations list covers the seven platforms enterprise teams cite most often when they are asked which agent framework they are standardizing on, and it is the broadest portable governance surface that any vendor has put on a single product page in 2026 to date.

The pattern that enterprise AI governance teams should be tracking for 2026 is the move from per-platform integration to portable policy that follows the agent, and Netzilo's announcement is the clearest example of that pattern shipping in production this week. The primary source for the integration list, the CEO quote, and the runtime graph detail is the Help Net Security writeup of the announcement, and the most relevant prior AIntelligenceHub coverage is the Snyk Evo ADS launch for the coding-agent governance angle. The enterprise AI governance checklist for 2026 is the right landing page for teams turning this announcement into a buying decision, and it pairs naturally with the runtime graph, the Governance-as-Code primitives, and the portable policy claim that AIDR is built around.