How Axonius Leverages AI with AI Companion & AI Workforce (Formerly AI Agent in EA) - Subject to GA

Feature Overview

Axonius AI Companion & AI Workforce are productivity and support capabilities designed to act as in-product technical assistants. To ensure safety and regulatory compliance, the AI Companion & AI Workforce operate strictly as productivity tools.

While the AI Companion helps users analyze data, find answers across UIs, and draft precise action parameters directly within the platform dashboard, the AI Workforce executes actions under a human-supervised model and requires explicit user confirmation by default. While admins can enable automated execution for standard actions via platform settings, high-risk tasks—such as user and role management—always require human approval and cannot be automated.

Customers’ Control, Accessibility, and Settings

Activation Options & Functional Impact. For legacy customers, AI Companion & AI Workforce are disabled by default but can be enabled by account administrators via an in-product toggle. For new customers, these features are enabled by default during the POV and can be managed through the same toggle. If disabled, the feature interface is completely hidden from view.

Deployment and AI models. The AI Companion & AI Workforce rely on Anthropic and are available only to SaaS customer instances. To ensure maximum data privacy, isolation, and control, every customer is provisioned their own private AI LLM instance that runs entirely within their own dedicated environment.

Data Handling. To deliver actionable insights, the AI Companion & AI Workforce require access to customer tenant data; masking, pseudonymizing, or fully de-identifying this information would restrict the product's core utility. Customer data is continuously maintained within a logically isolated environment unique to each instance. Axonius strictly prohibits the pooling, combining, or cross-tenant aggregation of customer data.
Customer data is stored in the same location as the primary Axonius Platform instance. For select regions (ap-southeast-1, ap-northeast-2, sa-east-1, and me-central-1), localized hosting is unavailable; in these cases, the underlying infrastructure hosts processing services in external geographic regions optimized for proximity and routing performance.

Training and Optimization

The AI Companion & AI Workforce do not use customer or user-provided data to train or retrain the global models. Instance-level fine-tuning is a core requirement of these functionalities; all resulting fine-tuned models and associated data are strictly isolated per tenant. The AI Companion & Workforce automatically learn from customer data to optimize results exclusively for each specific customer organization.

Transparency, Explainability, Human Oversight & Action Automation

The AI Companion interface includes a persistent footer that clearly disclaims the AI interaction. From this footer, a [Learn More] link directs users to additional details on the Axonius documentation site. To ensure seamless escalation, an obvious path to a human agent is positioned immediately adjacent to the chat input field, providing direct entry to the Axonius Support Portal.

For complete transparency, users can directly ask the AI Companion to explain its reasoning or click the [Show Thinking] button in the chat interface to review its step-by-step logic.

By default, the AI Workforce operates under a human-supervised model. It drafts parameters and suggests actions, but requires explicit approval via the "Approve this action" button before executing. Any modification to data must be confirmed by a user before it is finalized. While admins can enable automated execution for standard tasks via settings, high-risk actions—such as user and role management—can never be automated and always require human approval.
Admins retain full control: they can instantly disable the AI Workforce, revoke execution privileges in settings, or instruct the AI to roll back an action. Every action is logged and tagged to indicate whether it was initiated by a human or the AI. Ultimately, the AI Workforce never engages in unsupervised automated decision-making.

Accuracy, Reliability, Security and Integrity

The AI models build their knowledge base using customer data, as validated by Axonius’ adapters. This localized anchoring fundamentally insulates them from external threat actors and malicious outside manipulation.
Users can initiate a new chat session at any time, which completely clears the local context and prevents conversational history from bleeding into the new interaction. Additionally, administrators have the ability to globally disable persistent memory/history between chat sessions.
To ensure structural integrity and output quality, multi-layered security protocols and operational guardrails run continuously across the data pipeline. System reliability and response accuracy are maintained through a combination of structured evaluation and ongoing monitoring.
Multi-layered security and operational guardrails are actively applied across the data pipeline to verify structural integrity at every stage.
We also maintain a robust, multi-tiered Quality Assurance (QA) process. Every new version or update to the AI mechanism is evaluated against hundreds of diverse test prompts spanning a wide range of scenarios. For each prompt, we compare the agent's response to an expected answer and assign a numerical score — creating a clear evaluation framework that shows whether a given change improved, maintained, or degraded performance. This process helps us catch regressions early and iterate quickly on any areas that fall short of our benchmark.
Reliability is further reinforced through a scheduled monitoring lifecycle. Following deployment, we conduct periodic performance and accuracy assessments to refine model outputs and keep pace with an evolving security landscape. Each test cycle covers hundreds of questions across diverse domains, and we run an equivalent process as an end-to-end sanity check of the entire feature.

Retention

During an active contract, interaction history is retained by default. Admins can configure their preferred auto-deletion schedule for chat history at any time via System Settings → AI Usage.
Upon contract termination, interaction history is retained for a standard period to support debugging and continuity, then automatically deleted.