Trust Center – Private AI

Compliance & Certifications

Independently verified, always current

Our certifications are maintained through continuous monitoring and monthly internal reviews. ISO 27001 and ISO 27701 were independently certified in December 2025. All documentation is available on request following a short NDA process.

🛡️

ISO 27001

Information Security Management System — independently certified December 2025. The gold standard for enterprise security controls.

Certified Dec 2025

🔒

ISO 27701

Privacy Information Management System — independently certified December 2025. Extends 27001 to cover personal data handling and privacy.

Certified Dec 2025

🇸🇬

MAS TRM

Monetary Authority of Singapore Technology Risk Management Guidelines — our security controls, infrastructure, and governance practices are aligned to MAS TRM requirements.

Aligned

🇦🇺

Essential Eight

Australian Cyber Security Centre’s Essential Eight Maturity Framework — aligned at Maturity Level 2, targeting the controls defined in the ACSC guidance.

Aligned — Targeting ML2

🇪🇺

GDPR

Full alignment with EU General Data Protection Regulation across data collection, processing, and retention.

Aligned

🌏

PDPA

Personal Data Protection Act compliance for customers operating across ASEAN markets.

Aligned

🏥

HIPAA Readiness

Controls architecture designed to support HIPAA requirements for health data — activated on client request.

Customer-Driven

📋

SOC 2

Security, availability, and confidentiality controls mapped to the SOC 2 Trust Services Criteria.

Controls Mapped

🧭

NIST CSF

Cybersecurity Framework mapping across Identify, Protect, Detect, Respond, and Recover functions.

Controls Mapped

🤖

ISO 23894

AI risk management guidance — applied to our agentic AI deployments and client solution governance.

Guidance Applied

🔭

ISO 42001

AI Management Systems standard — evaluation underway. Timeline driven by customer demand and regulatory developments.

Roadmap

🔧

CIS Benchmarks

Centre for Internet Security hardening benchmarks applied to infrastructure configuration where applicable across our AWS environment.

Aligned

Security Architecture

Defence in depth — from infrastructure to application

PrivateAI runs on our own certified AWS organisational environment with layered security controls at every tier. Your data is hosted within our ISO 27001-certified infrastructure — not a shared public cloud tenancy. No shared tenancy. No shortcuts.

☁️

AWS Cloud Infrastructure

Deployed on dedicated AWS instances across multiple availability zones. Infrastructure-as-code ensures reproducible, auditable environments with no manual drift.

🔐

Encryption at Rest & In Transit

AES-256 encryption at rest for all data stores. TLS 1.3 enforced across all client connections. Keys managed through AWS KMS with rotation policies.

👤

Identity & Access Management

Role-based access control (RBAC) enforced across all systems. MFA required for all administrative access. Principle of least privilege applied throughout.

🕵️

Monitoring & Threat Detection

Continuous log aggregation and anomaly detection via AWS CloudWatch and GuardDuty. Alerts routed to a 24/7 security response channel.

🧪

Vulnerability Management

Automated dependency scanning in all CI/CD pipelines. Periodic third-party penetration testing. Critical patches applied within 24 hours of disclosure.

🚨

Incident Response

Documented IR playbooks aligned to NIST SP 800-61. Affected clients notified within 72 hours of a confirmed incident, per GDPR requirements.

🏢

Network Isolation

Client environments deployed in isolated VPCs. No cross-tenant data access possible by design. Private networking between services with no public exposure.

💾

Backup & Recovery

Automated daily backups with 30-day retention. Cross-region replication for enterprise clients. RTO and RPO targets defined per service tier in client SLAs.

🧑‍💼

Personnel Security

Background checks for all staff with access to production systems. Security awareness training completed annually. Contractors bound by NDA and access-limited by default.

AI & Data Protection

How we govern AI processing of your data

Our AI and LLM processing is governed by privacy-by-design principles. We apply data minimisation, access control, logging and retention controls, and conduct vendor risk reviews before any new AI subprocessor is introduced.

✂️

Data Minimisation & Redaction-First

We apply a redaction-first approach to AI processing — only the minimum data required to answer a query is passed to a model. Sensitive fields are masked or excluded by default unless explicitly scoped.

📋

Controlled Logging & Retention

AI query logs are retained only as long as necessary for audit and debugging purposes, under documented retention controls. Clients can request log retention periods specific to their compliance requirements.

🔍

Vendor & Subprocessor Risk Assessment

Every AI model provider or third-party service that may process client data is assessed before use — covering data handling practices, contractual protections, and certification status. We do not onboard AI vendors without a documented risk review.

📝

Honest Claims, Not Marketing Language

We do not publish blanket “zero retention”, “no training”, or “EU-only” claims unless these are contractually evidenced for your specific service scope. Ask us what applies to your deployment and we will confirm it in writing.

🏠

Private Deployment Option

For clients requiring full model isolation, PrivateAI can deploy a dedicated AI model within our certified AWS organisational environment — fully isolated from other clients, with no data leaving our secure perimeter for inference. ISO 27001 and ISO 27701 controls apply to this environment. Available on Pro+ and Enterprise plans.

🧾

Traceable & Auditable Outputs

AI-generated outputs in our finance and compliance agents include citations and source traceability — designed to support audit preparation and compliance workflows, not replace human review.

Privacy & Data Handling

Your data stays yours

We operate on a strict data minimisation and purpose-limitation basis. Client data is never used to train models or shared with third parties without explicit consent.

🎯

Purpose Limitation

Data collected is used only for the purposes explicitly agreed with the client. No secondary use without separate consent.

✂️

Data Minimisation

We collect and retain only the minimum data necessary to deliver the contracted service. No speculative data collection.

🚫

No Model Training

Client data is never used to train, fine-tune, or evaluate AI models — ours or any third party’s. This is a contractual commitment.

🌍

Data Residency

Client data is hosted in the agreed AWS region. Cross-border transfers are governed by SCCs or equivalent mechanisms.

📝

DPA Available

A standard Data Processing Agreement is available for all clients. Custom DPA terms are negotiated for enterprise engagements.

👁️

Transparency Logs

Clients can request access logs showing who accessed their data, when, and for what purpose, at any time.

📂

Right of Access

You can request a full export of all personal data we hold about you or your organisation within 30 days.

✏️

Right to Rectification

Inaccurate personal data will be corrected promptly upon request. Contact [email protected] to raise a correction.

🗑️

Right to Erasure

You may request deletion of your personal data. Requests are fulfilled within 30 days subject to legal retention obligations.

⛔

Right to Object

You may object to processing of your data for any purpose where we rely on legitimate interests as our legal basis.

Submit a data rights request to [email protected]. We will acknowledge within 48 hours and fulfil within the statutory period.

Data Category	Retention Period	Basis	Disposal Method
Client operational data	Duration of contract + 12 months	Contractual obligation	Secure deletion / AWS data wipe
Audit & access logs	24 months	Legal / regulatory	Encrypted archive then deletion
Contact & account data	Contract term + 36 months	Legitimate interest	Anonymised or deleted on request
Incident records	5 years	Legal obligation	Archived, then securely deleted
Financial records	7 years	Tax / regulatory	Per statutory requirements

Subprocessors

Third parties that process client data

A subprocessor is a third party that processes personal data on our behalf as part of delivering the PrivateAI service. We maintain this list under GDPR Article 28. Clients are notified of any additions 30 days in advance, with the right to object. Self-hosted open-source tools running entirely within your AWS environment are not subprocessors — the processor relationship for those is with AWS.

Subprocessor	Purpose	Data Location	Engagement	Certification
Amazon Web Services	Cloud infrastructure & hosting — all client environments	Client-selected region (default: ap-southeast-2 Sydney)	Always	ISO 27001 SOC 2
Anthropic	AI model inference via API — when client workloads use Claude models	US / EU (configurable via enterprise agreement)	Conditional	Zero-retention API
Google Cloud	AI model inference or storage — when client workloads use Google models or GCS	Client-selected region	Conditional	ISO 27001 SOC 2
Microsoft Azure	AI model inference or storage — when client workloads use Azure OpenAI or Azure services	Client-selected region	Conditional	ISO 27001 SOC 2
Vast.ai	GPU compute for on-demand model inference — when client workloads require burst GPU capacity	Variable (region specified at job submission)	Conditional	Contractual DPA

Conditional means the subprocessor is only engaged when a client’s specific use case requires it — it is not active for all deployments. Last updated: April 2026. Contact [email protected] to request the full subprocessor register or to be notified of changes.

Common Questions

Security & compliance FAQ

Questions we’re asked in every enterprise sales process — answered up front so your security team can self-serve.

No. Client data is never used to train, fine-tune, or benchmark any AI model — ours or a third party’s. When we engage hosted model APIs (e.g. Anthropic, Azure OpenAI), we do so under enterprise agreements with data processing controls, and we conduct vendor risk assessments before use. We do not make blanket “zero retention” or “no training” claims unless these are contractually evidenced for your specific service scope — ask us and we will confirm in writing what applies to your deployment.

All client data is hosted in your nominated AWS region, specified at contract signing. Australian clients default to ap-southeast-2 (Sydney). We do not move data across regions without written consent. Data residency is documented in the DPA.

Production data access is restricted to a named subset of senior engineers on a break-glass basis. All access events are logged and auditable. Clients can request an access log at any time. We operate on the principle of least privilege — no standing access to client production environments.

Yes. Certificates are shared upon a signed NDA or as part of a formal vendor assessment process. Submit a request via [email protected] and our team will respond within two business days.

Annual third-party penetration testing is conducted on the PrivateAI platform. Executive summaries and remediation reports are available to enterprise clients under NDA. Contact [email protected] to initiate a request.

Our IR plan requires notification to affected clients within 72 hours of a confirmed breach, consistent with GDPR Article 33. Notifications include: what data was affected, how, what we’ve done to contain it, and recommended actions for you. A detailed post-incident report follows within 30 days.

Yes. All team members — regardless of location — are bound by our information security policy, complete annual security awareness training, and operate under NDAs. Access is provisioned on a role and project basis with no standing production access by default.

Yes. Send your questionnaire to [email protected]. Standard questionnaires (SIG Lite, CAIQ, custom) are typically turned around within 5 business days. We maintain pre-completed responses to common frameworks to accelerate your review.

The Australian Cyber Security Centre’s Essential Eight Maturity Model defines four levels (ML0–ML3). We have aligned our controls to the requirements defined at Maturity Level 2 — covering patching, application hardening, MFA, and restricted admin privileges to ML2 standards. We have not yet undergone a formal third-party maturity assessment. “Targeting ML2” is the honest and accurate way to represent our current posture.

MAS TRM (Monetary Authority of Singapore Technology Risk Management Guidelines) is a regulatory guideline, not a certifiable standard. Alignment means our infrastructure and security controls are designed to meet MAS TRM requirements — covering technology risk governance, system resilience, access controls, and incident management. Singapore-regulated clients can request our MAS TRM control mapping document via [email protected].

Security & compliance
you can verify.

Accelerate your vendor assessment

Security & complianceyou can verify.

Accelerate your vendor assessment

Security & compliance
you can verify.