top of page

Legal and Trust Centre

 

At Elf Works Pty Ltd (ACN 678 538 086), we are dedicated to safeguarding the privacy and security of personal and sensitive information collected and processed through our platform-as-a-service (PaaS) software, which supports professionals and organisations with advisory, research, and knowledge-generation tasks.

 

Our platform uses two categories of Large Language Model (LLM). Sovereign models are language models hosted on Australian infrastructure, where processing takes place entirely onshore. Frontier models — ChatGPT, Claude, Gemini, and Grok — are operated by providers located outside Australia and are used where a task requires their additional reasoning capability. Direct client identifiers are removed before any request is sent to a frontier provider.

 

This Trust and Security page serves as a centralised resource outlining our commitments, practices, and measures to protect your data, ensure compliance, and build trust. We comply with the Privacy Act 1988 (Cth) and Australian Privacy Principles (APPs).

 

Our security practices are designed in alignment with ISO 27001 and SOC 2 control objectives, with formal certification planned as we scale.

Our Commitment to Trust

We prioritise transparency, data protection, and empowering your rights as a digital service user.

We do not allow any user data passed to Large Language Models (LLMs) to be used for training or retained beyond immediate processing needs — all data is handled ephemerally and deleted promptly after use, as enforced through contracts with providers.

We offer sovereign, onshore language model processing so that work involving client records and other sensitive information can be completed without that information leaving Australia.

We regularly review and update our practices to address emerging threats, regulatory changes, and feedback.

Our platform, data centres, and sovereign model inference are Australia-based. Where a task uses a frontier model operated offshore, the request is de-identified first and the transfer is made under a Data Processing Agreement with contractual safeguards including Standard Contractual Clauses.

 

Sovereign Language Models

Alongside the frontier LLM providers listed in our vendor list, we use sovereign language models hosted entirely within Australia.

  • Onshore inference. Sovereign model processing runs on ASIC-based infrastructure at Equinix SY5 in Sydney. Data does not leave Australian jurisdiction at any point during this processing.

  • Model type. Sovereign processing uses open-weight language models hosted by the provider on Australian infrastructure, rather than proprietary models operated from overseas.

  • Jurisdiction. Because the provider is Australian-incorporated with no foreign parent entity, data processed through it is not exposed to foreign legal process such as the US CLOUD Act.

  • Retention. Prompts are not stored, indexed, or replayed, and are not used to train or fine-tune any model. Data is processed in-memory and discarded immediately after inference.

  • When sovereign models are used. [Confirm: describe which tasks or data categories route to sovereign models by default — for example, work involving client records, tax file numbers, or other sensitive identifiers.]

 

De-identification before offshore processing

Where a task requires a frontier model, direct client identifiers — including tax file numbers [confirm the full field list] — are removed before the request is transmitted. The frontier provider receives only de-identified content, and identifiers are re-associated with the output within our Australian environment after the response is returned.

For provider-by-provider detail on what each LLM provider processes, where it is located, and its retention terms, see our vendor list at www.elfworks.ai/trust-centre/vendors.

Security Measures

We implement robust safeguards to protect data across our platform, development environment, and integrations:

  • Infrastructure Security: Hosted on Australian-based AWS infrastructure in Sydney, NSW, with ISO 27001-certified vendors. Sovereign model inference is performed on Australian infrastructure in Sydney. We use firewalls, intrusion detection systems, real-time logging, monitoring, and quarterly-tested disaster recovery protocols with AES-256 encrypted backups.

  • Data Encryption: All data in transit is encrypted using TLS 1.3, and data at rest uses AES-256 encryption.

  • Access Controls: Guided by principles of Least Privilege (PoLP), Role-Based Access Control (RBAC), Separation of Duties (SoD), and Need-to-Know. Access levels include:

    • For the Platform: System Admin (full access, including API integrations), Advisor (create/export advice using sovereign and frontier LLMs), Professional User (view-only, no LLM interaction).

    • For Development Environment (AWS-hosted, including Streamlit.io and MongoDB): System Admin (full access), Developer (project-specific access to tools and API credentials).

    Multi-factor authentication (MFA/2FA) is required for all access, using Microsoft for the platform (extracting only email and username), AWS IAM for development and MongoDB, and secure mechanisms for API keys stored in AWS Secrets Manager. Credentials for sovereign and frontier model providers are held in AWS Secrets Manager on the same basis.

  • Application Security: Network segmentation via AWS VPCs and firewalls limits lateral movement.

  • Monitoring and Logging: All access attempts (successful/failed) to the platform, development environment, MongoDB, and API credentials are logged with user ID, timestamp, and resource details. Logs are reviewed monthly, with real-time alerts for anomalies like failed logins or unauthorised access using AWS CloudTrail and MongoDB tools.

  • Physical Security: For AWS data centres, refer to the AWS Trust Center for details on physical controls. Sovereign inference infrastructure is located in a colocation facility operated to equivalent standards.

 

Compliance and Certifications

  • Regulatory Compliance: We comply with the Privacy Act 1988 (Cth), SOC 2 Type II (in alignment), ISO 27001, GDPR, and other standards. Data processing agreements (DPAs) are available to eligible users.

  • Employee Training: All employees receive annual training on access controls, phishing awareness, data protection, and secure handling of the platform, development environment and LLM API credentials. New hires complete onboarding within their first week.

 

Vendor Management and Data Sharing

We share data under strict controls with vetted third parties for authorised purposes only:

  • Cloud hosting/infrastructure (e.g., AWS).

  • Sovereign AI inference (SCX.ai / SouthernCrossAI, Australia).

  • Frontier LLM providers (OpenAI, Anthropic, Google, xAI), which receive de-identified content only.

  • Payment processors (e.g., Stripe, which handles/stores payment details; see Stripe Trust Center).

  • Government authorities when legally required.

 

All vendors comply with our standards and are restricted via contracts. For the full vendor list, visit www.elfworks.ai/trust-centre/vendors.

 

This list is subject to updates as we add or change vendors. We notify users of material changes via our Privacy Policy at www.elfworks.ai/trust-centre/privacy. If you have questions or wish to exercise your data rights, please contact us at info@elfworks.ai.

 

We recommend reviewing each vendor's privacy policy for additional insights into their practices.

Incident Response

We have a documented incident response plan, reviewed annually, to handle breaches or unauthorised access swiftly:

  1. Contain the incident.

  2. Notify affected users and the OAIC (if required) within 72 hours for confirmed breaches.

  3. Provide remedial support.

  4. Conduct root-cause analysis and implement improvements.

Suspected incidents trigger immediate investigation by our technical team. We participate in responsible disclosure and encourage vulnerability reporting via info@elfworks.ai.

International Data Transfers

Our platform infrastructure, data storage, and sovereign model inference are located in Australia.

Where a task uses a frontier LLM (OpenAI, Anthropic, Google, or xAI), the request is processed by that provider outside Australia, primarily in the United States. Before any such transfer:

  • Direct client identifiers are removed, so the provider receives de-identified content.

  • The transfer is made under an executed Data Processing Agreement including Standard Contractual Clauses.

  • The provider is contractually prohibited from retaining the content beyond inference or using it for model training.

Users who require that no data be processed offshore in any form can [confirm: describe the available option — for example, restricting their firm's workspace to sovereign models only, if that setting exists].

Future expansions will include safeguards such as Binding Corporate Rules or Standard Contractual Clauses.

Continuous Improvement

Policies are reviewed annually or after significant changes. Employees and customers can submit feedback on practices via info@elfworks.ai.

 

Contact Us

For questions about our trust and security practices, contact our Data Protection Officer or Technical Team at info@elfworks.ai. We respond within 30 days.

This page was last updated on 24 August 2026. We notify users of significant changes via email, with a summary and 30 days' notice before enforcement.

Legal and Trust Center Documents

bottom of page