AI security for Atera’s autonomous AI agents

Atera is committed to developing and using AI in a secure, ethical, and trustworthy way. The guiding principles behind every feature of Robin and AI Copilot, our autonomous AI agents for IT, align with global standards such as Microsoft’s Responsible AI Principles, the NIST AI Risk Management Framework, and the ISO/IEC 42001 AI Management System.

Important

Your prompts (inputs) and completions (outputs), your embeddings, and your training data:

are NOT available to other customers.
are NOT used to improve OpenAI models.
are NOT used to train, retrain, or improve Azure OpenAI Service foundation models.

The Azure OpenAI Service is operated by Microsoft as an Azure service; Microsoft hosts the OpenAI models in Microsoft’s Azure environment and the Service does NOT interact with any services operated by OpenAI (e.g. ChatGPT, or the OpenAI API).

Microsoft Security Page

Setting the standard for responsible AI in IT

The widespread adoption of generative and autonomous AI systems demands that the tech industry adopt strong internal security practices to safeguard users, systems, and data. As an innovation-first company, Atera aims to not only keep pace with leading companies in responsible AI development, but also set a new standard for the IT industry by embedding robust, transparent security principles into our AI.

Data privacy

Your data is exclusively yours. By hosting on Microsoft Azure OpenAI, Atera provides enterprise-grade security while ensuring your information is not used to train external or internal models. All interactions and data remain strictly isolated within your own tenant, guaranteeing complete confidentiality.

Human-centric design

You are always in control. Atera’s AI agents are designed to augment your team, with administrators governing all behavior through custom policies. Every AI-generated suggestion requires mandatory human approval before it is activated, ensuring that technology always serves your expert judgment.

Secure-by-design

Atera provides multi-layer security through a secure-by-design framework. Our AI agents take only pre-vetted, low-risk actions, while administrators keep full control over enabling custom scripts or advanced functions. Always-on content filters and a zero-trust architecture reinforce this proactive stance, ensuring strict data isolation.

Transparency and accountability

Atera eliminates AI’s “black box” by providing complete visibility and control. Every autonomous action taken by Robin is logged with the same traceability as a human technician, and all AI decisions are recorded as internal ticket notes, giving your team a clear understanding of not just what the AI did, but why.

The Atera Trustworthy AI Assurance Framework (TAAF)

Atera has developed the TAAF to ensure our AI use remains secure, ethical, and trustworthy. This framework defines
the operational principles guiding every feature of Robin and AI Copilot, our autonomous AI agents for IT.

Atera has developed the TAAF to ensure our AI use remains secure, ethical, and trustworthy. This framework defines the operational principles guiding every feature of Robin and AI Copilot, our autonomous AI agents for IT.

All the AI models used for Robin are hosted on Microsoft Azure OpenAI, ensuring enterprise-grade security and data sovereignty. No data is shared with OpenAI or used to train models. Prompts and interactions are never logged or reused, and your data always remains isolated from other tenants.

Atera provides hundreds of built-in actions and diagnostics that are developed, maintained, and rigorously tested by our engineering team. Each action undergoes thorough review and security evaluation to ensure safe operation within your environment. By default, only those actions with a low risk score of 2 or below are enabled. These built-in capabilities address a wide range of IT use cases, including user onboarding, device diagnostics, script automation, and policy enforcement. One example is the password reset action, which is enabled by default and secured using SMS-based two-factor authentication.In cases where the default behavior needs to be adjusted to suit specific operational requirements, admins can define Custom Instructions to override default settings. Custom scripts and third-party integrations, however, are never enabled by default, but are fully admin-controlled to maintain a secure and predictable execution.

Robin includes a real-time content filtering engine that blocks all non-IT topics as well as harmful content such as hate speech, violence, adult content. These filters are always on and cannot be disabled, ensuring interactions stay compliant and professional.

Each tenant’s knowledge base, scripts, logs, and configurations are stored in a dedicated database. For MSPs, data is also segmented per end-customer, ensuring no cross-tenant visibility. This multi-tenant security model is aligned with zero-trust data boundaries.

Every action taken by Robin is recorded in Atera’s Audit Log, just like those performed by human technicians. Each log entry clearly identifies “Robin” as the executing entity, ensuring there is no ambiguity about who or what performed the task.This level of transparency allows administrators to track every change, action, or response generated by the AI agent, ensuring agent actions meet the same rigorous standards of traceability and accountability as any human users.

Every interaction between end users and Robin is automatically logged as a ticket in Atera, with all AI decisions recorded as internal comments within the same ticket. These comments are only visible to technicians and admins, providing clear insight into the AI’s reasoning and behavior. This allows IT teams to review context at any time, ensuring transparency, oversight, and confidence in the agent’s actions.

Robin behavior is governed by Custom Instructions, i.e., account-level policies written by IT administrators to ensure they remain in full control of the agent’s scope and behavior. These instructions define how the agent acts, what it is allowed to do, and under what conditions it escalates or responds.

AI generated content, such as knowledge base articles and script suggestions, is optional, and requires manual human admin approval before they’re published or executed. No outputs are ever implemented without review.

The TAAF matrix

Principle	What it means for you	Default behavior
Privacy-respecting AI	Your information is not used to train external or internal models	Enforced
Admin-governed autonomy	Control Robin behavior via Custom Instructions	Fully configurable
Risk-rated action control	Only low-risk actions enabled by default	Max risk level 2 by default
Password reset with 2FA	SMS verification before execution	Enabled
Optional AI recommendations	Optional and reviewed before use	Off by default
Harmful content filtering	Filters block non-IT and harmful inputs	Always on
Tenant/customer data isolation	Segregated databases per tenant/customer	Always enforced
Ethical AI commitment	Built on Microsoft Responsible AI standards	Enforced
Built on Microsoft Responsible AI standards	All actions logged as ‘Robin’ technician	Always on
Conversation and decision logging	All user-AI chats and reasoning logged as tickets/comments	Always on

Want to learn more?

Our security team is happy to provide architecture diagrams, risk assessments, and compliance checklists.
Contact us at [email protected]

Visit us at trust.atera.com