Security Information Guide: CISO and Vendor Risk Answers
What CISOs and vendor risk teams need to know about Audition AI: where your data lives, how AI and agent actions are governed, who can access your deployment, and how it stays up to date.
Last updated: November 11, 2026
Audition AI is an enterprise AI platform that runs in your organization's own Azure subscription. All of your content (chats, documents, uploaded files, search indexes and audit records) is stored in Azure Storage and Azure SQL that you own, and your Azure Entra ID controls who can sign in and what they can do.
Audition AI is off by default. Out of the box, connectors, data sources and capabilities are disabled, and each one is turned on deliberately by your administrators. That means policy enforcement, from model access to agent guardrails to retention, is something you configure. Our team offers configuration guidance and configuration services to help.
Our security model follows a shared responsibility approach: Azure's security infrastructure, application-level controls from Audition AI, and the policies your team sets.
Shared Responsibility Model
Audition AI and its customers share security responsibilities to ensure a comprehensive security posture.
- Secure Application Development: Following secure SDLC practices with code reviews and penetration testing
- Identity Management Integration: Seamless integration with Azure Entra ID
- Encryption: Implementing encryption for data in transit and at rest
- Governance Controls: Providing the controls your admins use to govern AI, including model access, approvals, Guardian rules and retention
- Regular Updates: Shipping regular security and dependency updates
- Security Policy Configuration: Setting organizational security policies within Azure
- Access Control: Managing user access and permissions via Azure Entra ID
- AI Policy Configuration: Choosing models, enabling connectors, and configuring Guardian rules and retention
- Compliance Monitoring: Ensuring compliance within your Azure tenant
- Update Cadence: Choosing a release cycle for applying updates
Data Location & Protection
All customer content (chats, documents, uploaded files, search indexes and audit records) is stored in Azure Storage and Azure SQL in your own Azure subscription, in the region you deploy to. You choose where your deployment runs to meet your data residency requirements.
Audition AI uses dedicated Azure resources in your subscription:
- Azure Blob Storage for documents, files and conversation records
- Azure SQL Databases for structured data
All data is encrypted both in transit and at rest:
- At Rest: AES-256 encryption for all stored data, through Azure Storage service-side encryption and Azure SQL transparent data encryption, both on by default
- In Transit: TLS 1.2+ for all network communications
- Keys: encryption keys are Azure-managed. Customer-managed keys (CMK) aren't supported today. If you need them, tell us: we're collecting customer feedback.
By default, deployments share support telemetry with Audition AI so we can provide proactive support:
- Licensing data, including the list of licensed users
- Usage counts
- Configuration
- Error data
Support telemetry never includes chat or document content. You can turn it off so that no data leaves your environment. Most customers keep it on.
We regularly review telemetry to confirm it contains no customer content. If content ever slipped in, it would be a mistake, not a design choice, and we would fix it as a bug.
Audition AI support can only access the application after one of your administrators approves a request. Your administrators can revoke that access instantly, and it also ends when the application restarts.
Private deployments: for customers who require it, our team configures private networking after installation and removes Audition AI's database access as part of that configuration.
Audition AI is a tool your organization uses to work with other systems that process your data:
- Microsoft, through your Azure subscription, remains your primary processor
- Third-party services you enable, such as the model providers your administrators choose
AI & Model Security
Every AI capability is off until your administrators turn it on. These are the controls they use, and our team can help you configure them.
Available to Audition AI Enterprise customers
Guardian watches every step an AI agent takes: what people send, what tools return, what the agent tries to do, and what the AI says. Administrator rules decide what happens next: allow the step, deny it, require a person to approve it, or stop the agent mid-turn. Rules run in a sandbox with time, memory and AI-call limits.
Ready-made rule templates include:
- Prompt injection: checks tool results for injected instructions by phrase matching, with an optional AI check
- Secrets and passwords: pattern-based detection
- Personal data: Social Security numbers, payment card numbers (with a checksum test) and your own keywords, across user input, tool results, AI output and tool calls
- Email: send only to approved domains, and limit bulk email
- Safety rules for SQL, PowerShell and computer use
- Regulated advice
Like everything on the platform, your administrators enable Guardian and choose its rules.
Explore Guardian- You choose the model endpoints. Admins configure each model deployment's endpoint and key, for example Azure OpenAI or Azure AI Foundry (including Claude) in your own subscription. The deployment can create the Azure OpenAI resource in your subscription for you.
- Model access per user. Admins can restrict each model to specific users.
- Sensitivity-aligned routing, admin-configured. Admins can turn off user model selection, bind each approved model to a persona, and scope that persona's SharePoint knowledge sources with metadata filters, including Microsoft Purview sensitivity labels. See the Data Sovereignty Rules Framework.
- Azure content filtering. When the deployment creates your Azure OpenAI resource, it applies blocking content filters for hate, sexual content, violence and self-harm to prompts and completions.
- Training. Under Microsoft's terms, Azure OpenAI does not use your prompts or completions to train models. See Microsoft's data, privacy and security documentation.
- We don't train on your data. Audition AI does not train models on your company data, and nobody at Audition AI or related companies has access to data that would be useful for model training. We use support telemetry to tune our agents and automations, which is not model training.
- Live Microsoft 365 reads run as the user. Live reads of SharePoint, OneDrive and Outlook through Microsoft Graph use the signed-in user's own delegated token, so they return only what that user can already access.
- Search results are re-checked against live permissions. Every SharePoint and Egnyte result from enterprise search is re-checked against the signed-in user's current permissions before the AI sees it. If access can't be confirmed, the result is withheld. OneDrive, Outlook and Teams results are scoped to the user.
- Approval for sensitive actions. Actions with side effects, such as sending, deleting or writing, require a person to approve them before they run. Admins can deny specific actions entirely.
- Agent limits. Admins can cap how many tool-call rounds an assistant can make per request.
Identity & Access Management
Full integration with Azure Entra ID for seamless enterprise authentication. Users authenticate using their existing organizational credentials with no separate password management required.
Customer-Configurable
Audition AI does not enforce MFA directly. Security best practices for MFA are enforced through Azure Entra ID policies configured by your organization.
Audition AI separates duties across roles. Nobody has a role until an administrator assigns one:
Standard User
Access limited to their own conversations and the resources assigned to them.
Compliance
Oversees rules and reviews conversation history across the organization.
Admin
Manages all access and connections, but can't change rules or view anyone else's conversation history.
Persona Admin
Configures personas, with no access to compliance, rules or access management.
Developer
Access to advanced developer tools.
Sidekick
Can use Sidekick, the local desktop companion.
Autonomous Agents
A role with its own rules, just for bots and digital coworkers.
Audition AI works with Microsoft Entra Conditional Access and Microsoft Defender for Cloud Apps session controls, so your administrators can decide what users can do with content based on the device they're using. For example:
- On company-owned, managed computers, users can copy and paste as normal
- On personal devices, admins can block file downloads, copy, paste and print
- Download controls can use Microsoft Purview sensitivity labels to block only labeled files
- Admins can require step-up authentication before a sensitive action
These controls are off until your administrators configure them in your own Entra and Defender tenants. Audition AI mobile apps aren't covered by these session controls yet.
Set up Conditional Access App ControlAudition AI ships with nothing on. Access starts at zero and is added only as each job requires:
- No access by default. No user, connector, data source or model has access until an administrator grants it.
- Least privilege. Admins add access per user, model and persona, only as needed to do the job.
- Verified identity. Every user signs in through your Azure Entra ID, and your Conditional Access policies can require device health and continuous verification.
- Everything recorded. Every conversation, data access and agent action is captured in the thread log.
Network & Application Security
Audition AI follows a multi-layered security model, leveraging Azure's capabilities with optional customer-configurable protections.
Network Security
- Secure API Gateway
Secure, scalable API access.
- Virtual Network Integration
VNet integration, with VNet rules on Azure SQL and Azure Storage.
- Private Networking on Request
For customers who require it, our team configures private networking after installation.
- ○DDoS Protection
Customers can enable Azure DDoS Protection on the virtual network in their Azure subscription.
- ○Web Application Firewall (WAF)
Not enforced by Audition AI. Customers can place Azure Web Application Firewall in front of the application, for example through Azure Application Gateway or Azure Front Door.
- ○Conditional Access
Customers can enforce access policies via Azure Entra ID based on their security requirements, including session-level controls via Microsoft Defender for Cloud Apps Conditional Access App Control.
Application Security
- Secure Software Development Lifecycle (SDLC)
Comprehensive code reviews and regular penetration testing.
- OWASP Top 10 Assessments
Regular vulnerability assessments against the OWASP Top 10 security risks.
- Secure API Access
OAuth 2.0 and token-based authentication for all API endpoints.
Logging, Monitoring & Incident Response
- Azure Log Analytics
Real-time monitoring and security event tracking integrated with Azure Monitor.
- Route to Your SIEM
Your Azure Monitor and Log Analytics data can be routed to Microsoft Sentinel or the SIEM of your choice.
- Immutable Audit Logs
Every conversation is recorded as an append-only thread log: written once, never edited, and never deleted by the application except under a retention policy your administrators set. What we mean by immutable.
- Automated threat detection and alerts
- Incident response procedures aligned with NIST 800-61 best practices
- Incident response plans tested with tabletop exercises
- Notification within 72 hours if a confirmed security incident affects your deployment or your data
Report a vulnerability: security@audition-ai.com. Our contact details are also published at /.well-known/security.txt.
Data Retention & Deletion
- You own retention. Your administrators (Administrator or ComplianceAdmin) can set a retention policy from 1 to 3,650 days that automatically purges conversations and their files. The period is measured from each conversation's most recent activity.
- Nothing is purged until you decide. Retention is off until an administrator turns it on, and individual conversations can be marked to keep indefinitely.
- Your own Azure tooling. All data sits in your Azure Storage and Azure SQL, so you can back up, export, retain or delete it with your own Azure tooling and policies, independent of Audition AI.
- When your contract ends. Your data stays in your Azure tenant, and you decide what to do with it. Data Audition AI collects automatically, such as support telemetry, expires through normal retention cycles of up to 7 years. Business records such as email, contacts, CRM and invoice data are kept as needed for business purposes. Our data protection agreement defines the details.
Frameworks Your Deployment Can Align To
Audition AI's architecture and deployment model help organizations align with these frameworks. These are not certifications held by Audition AI.
Because Audition AI runs in your own tenant, it falls under your certifications, audits and inspections. Your auditors can examine it directly, with no help needed from the Audition AI team.
- SOC 2
Security, availability and confidentiality controls
- ISO 27001
Information security management systems
- GDPR & CCPA
Data privacy
- HIPAA
Healthcare data protection, where applicable
- Azure compliance offerings
Microsoft's certifications and attestations for the Azure services you deploy on
Important: Actual compliance depends on your organization's complete security posture and implementation.
Business Continuity & Disaster Recovery
- Database: Azure SQL automatic backups with point-in-time restore, with transaction log backups about every 10 minutes
- Application: Azure App Service automatic backups
- Files: deleted files in Azure Storage can be recovered during a soft-delete window, and you can add Azure Backup for longer-term file backups
- ○Optional: Multi-region redundancy available for disaster recovery
- ○Optional: Hot-hot with auto-scaling. An optional upgrade for uninterrupted service: run active-active with auto-scaling. It's an Azure configuration, and our team can set it up.
- ○Optional: Availability Zones. Deployment across multiple Azure Availability Zones is supported as an optional upgrade.
When the Audition AI team supports your recovery, these are the objectives we promise:
Recovery Time Objective (RTO)
< 2 hours
Recovery Point Objective (RPO)
< 15 minutes
Need tighter objectives? Because Audition AI runs entirely in your own Azure subscription, your MSP or in-house team can achieve much tighter recovery objectives using built-in Azure features and configurable settings, such as zone redundancy, geo-replication, database failover groups, storage backups and hot-hot auto-scaling.
Updates & Patching
- Regular updates. Audition AI ships regular updates, daily and weekly, in response to bugs, security threats, and library and dependency updates.
- You choose when. You decide when updates are applied to your deployment by selecting a release cycle.
- Critical vulnerabilities patched within 24 hours. For recent high-profile third-party vulnerabilities, we shipped patches in under an hour.
How We Run Audition AI
The sections above cover the product. This one covers the company behind it.
- A named security lead owns our security program
- Written security policies, reviewed every year
- A risk assessment at least once a year
- An inventory of our data, systems and devices, with data classification
- A business continuity plan, tested regularly
- Background checks for new hires
- Security awareness training every year
- Confidentiality agreements for all staff and contractors
- Multi-factor authentication on every staff account
- Managed company devices with disk encryption
- Audition AI access to your deployment is limited to support, and private deployments remove it entirely
- Access reviews at least quarterly, and whenever staff change
- Automated dependency scanning flags vulnerable packages
- Every change is reviewed before it ships
- Releases are built and published only by our automated pipeline, never from a developer's machine
- Audition AI staff can't see your chat content
- Supports your GDPR and CCPA obligations: your data stays in your subscription, and you control retention, export and deletion
- Privacy questions: privacy@audition-ai.com
- Your deployment inherits the physical security of Microsoft's Azure data centers
- Our own infrastructure is kept under lock and key, monitored electronically, with electronic access control to both the building and the room
- No security breaches
- We carry insurance. For security reasons, we don't disclose carriers or coverage levels, publicly or privately.
Audition AI runs on Azure in your subscription, so it inherits Microsoft's sustainability commitments.
Questions CISOs Ask
All customer content (chats, documents, uploaded files, search indexes and audit records) is stored in Azure Storage and Azure SQL in your own Azure subscription.
Only if you allow support telemetry. By default, deployments share licensing data (including the list of licensed users), usage counts, configuration and error data with Audition AI so we can provide proactive support. It never includes chat or document content. You can turn support telemetry off so that no data leaves your environment.
You do. Your administrators can set a retention policy from 1 to 3,650 days, measured from each conversation's most recent activity. Until they do, nothing is purged. Individual conversations can be marked to keep indefinitely, and because all data sits in your own Azure subscription you can also back up, export or delete it with your own Azure tooling.
The models your administrators configure. Admins choose each model endpoint, such as Azure OpenAI or Azure AI Foundry in your own subscription, and can restrict each model to specific users or bind approved models to personas.
Actions with side effects, such as sending, deleting or writing, require a person to approve them before they run, and administrators can deny specific actions entirely. With Guardian, administrators can also write rules that allow, deny, ask a person, or stop an agent mid-turn.
Support access to the application requires one of your administrators to approve the request, and your administrators can revoke it instantly. For customers who require it, our team configures private networking after installation and removes Audition AI's database access.
Audition AI ships regular updates, daily and weekly, in response to bugs, security threats and library and dependency updates. You choose when updates are applied to your deployment by selecting a release cycle.
No. Audition AI does not train models on your company data, and nobody at Audition AI or related companies has access to data that would be useful for model training. We do use support telemetry to tune our agents and automations, which is not model training.
No. Audition AI has had no security breaches.
Your data stays in your own Azure tenant, and you choose what to do with it. Data Audition AI collects automatically, such as support telemetry, expires through normal retention cycles of up to 7 years. Business records such as email, contacts, CRM and invoice data are kept as needed for business purposes. Our data protection agreement defines the details.
If a confirmed security incident affects your deployment or your data, we notify you within 72 hours.
Audition AI is a tool your organization uses to work with other systems that process your data. Microsoft, through your Azure subscription, remains your primary processor, along with any third-party services your administrators choose to enable, such as model providers.
Related Resources
- Guardian →
Rules that allow, deny, ask a person or stop agent actions
- Immutability →
What our audit logs record, how we prove chain of custody, and how retention works
- Azure Entra Enterprise App Integration Guide →
Connect Audition AI to your Microsoft Entra ID
- Conditional Access App Control →
Copy, paste and download controls by device
- Data Sovereignty Rules Framework →
Route data to approved models by sensitivity
- Models →
The models your administrators can enable
- Privacy Policy →
How we handle personal information
Need Security Configuration Assistance?
Audition AI is off by default, and our team offers configuration guidance and configuration services to help you turn on exactly what your policies allow. Schedule a consultation to discuss your security requirements and compliance needs.