Purpose of this statement
This statement describes how Salsa Digital uses artificial intelligence (AI), how we govern that use, and what protections we apply.
We publish this statement because we work mainly with Australian Government agencies. Many of them are bound by the Digital Transformation Agency’s (DTA) Policy for the responsible use of AI in government (DTA Policy v2.0). Salsa Digital isn’t an Australian Government non-corporate Commonwealth entity and isn’t required to publish under that policy. But we adopt it as our internal standard, and publishing this statement is part of that.
This statement is a vendor transparency document. It doesn’t replace the AI transparency statement an Australian Government agency must publish about its own AI use. Where Salsa Digital builds or operates AI inside a government client’s systems, the client remains accountable for that AI use case under the DTA Policy. The client publishes their own statement covering it.
For a requirement-by-requirement view of how we support client compliance with the DTA Policy, see Salsa Digital and the DTA AI Policy: how we support client compliance.
What we mean by AI
In this statement, “AI” and “AI system” follow the definition in the DTA Policy, which uses the updated definition from the Organisation for Economic Co-operation and Development (OECD):
An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.
Source: OECD, updated definition of an AI system, 2024. Adopted by the DTA Policy v2.0.
This statement covers two kinds of AI use:
- Internal: AI used at Salsa Digital to support our staff in delivering services to our clients
- External: AI we build, configure or operate in our clients’ systems, where we support our clients’ own AI adoption
We hold both to the same standard and apply the same internal governance, described below.
Why we use AI
Salsa Digital uses AI to deliver higher-quality, faster and, where appropriate, more secure services to our government clients. We adopt AI deliberately and within set limits. We don’t block it, which encourages unsanctioned use. And we don’t accept it without controls, which creates risk.
We’re not religiously committed to AI. We treat it as a tool: useful when it adds value, set aside when it doesn’t. We use AI to speed up work it does well, such as drafting, summarising, code review and spotting anomalies in operational data. We leave AI out of work where the cost of being wrong is high, where the input data is unsuitable, or where a person can do the job faster.
We don’t use AI to make decisions on behalf of clients or members of the public. AI may help us analyse, draft or summarise. People who are accountable for the decisions make them.
How we use AI
Our current AI use falls within a small set of well-defined use cases, all in the corporate and enabling domain. Mapped to the DTA Policy’s classification of AI use:
- Workplace productivity: AI helps our staff draft documents, summarise meeting notes, draft and review code, generate test artefacts and do similar internal tasks. A qualified person reviews outputs before they leave Salsa Digital.
- Analytics for insights: AI helps our service desk and operational teams summarise tickets and operational logs, and find patterns that would otherwise take longer to find.
- Software engineering support: AI helps developers with code completion, refactoring suggestions, code review, test scaffolding and explaining unfamiliar code. All AI-assisted code goes through the same review and testing gates as code written without AI.
We don’t currently use AI in any domain that directly affects members of the public. We don’t operate any AI-enabled service channel for the public on Salsa Digital’s behalf.
Where we build or operate AI inside a client’s systems, the client’s use case classification applies. We help our clients classify, assess and govern those use cases under the DTA Policy. The client remains accountable for their use cases.
Public interaction
Our internal use of AI doesn’t interact directly with the public. The public isn’t significantly affected by our internal AI use without human review. A qualified Salsa Digital staff member reviews every AI-assisted output in a client deliverable before it’s delivered.
Where we build or operate AI inside a government client’s systems, the client’s own AI transparency statement covers public-facing AI use. We won’t deploy public-facing AI for a client without:
- The client’s written approval
- An impact assessment
- Arrangements for human oversight and contestability that meet the DTA Policy’s expectations
The client remains responsible for the public-facing part of the use case.
Where data is processed and how we handle it
To align with Australian Government data sovereignty policies, our default for client-facing AI work is to use Australian-resident AI platforms, with inference and storage in Australian regions. Our default platform for this work is hosted in Australian regions, under data handling terms that prohibit using customer data to train models. For AI in our own internal work, such as drafting documents, we use an AI capability in our productivity suite that stays within our existing tenant boundary.
For AI tools that don’t meet these defaults, we apply extra controls before approval, including:
- Manual de-identification of personal information before it’s submitted to the tool
- Vendor settings such as Zero Data Retention, where available
- Restricting the tool to specific use cases and data classes
- Documented client acceptance for any cross-border processing of client data
We don’t allow personal AI accounts or free-tier consumer AI services for any Salsa Digital work, or with any non-public Salsa Digital or client data. We use approved enterprise subscriptions with documented data handling arrangements.
Our internal rules ban secrets in prompts. They also ban sensitive client information from going into any AI tool that isn’t specifically approved for it. We use the Protective Security Policy Framework (PSPF) as our guide to what information may go where, including PSPF Policy Advisory 001-2025 on using generative AI with OFFICIAL information.
How we manage risks and protect against negative impacts
We govern AI under an internal AI governance framework built around the three pillars of the DTA Policy v2.0:
- Strategy and oversight
- Preparedness and operations
- AI use case impact assessment
Internal policies and procedures support the framework, along with a register of approved AI use cases and a list of approved AI tools.
Specific measures we apply:
- Human review of AI outputs. A qualified person reviews every AI output used in a Salsa Digital deliverable before it leaves Salsa Digital. AI outputs are drafts, not finished work.
- Mandatory annual training. All staff and contractors who use AI must complete training on this statement, our internal AI rules, data classification and incident reporting. Training is refreshed every year.
- Approved tools only. Staff and contractors may only use AI tools on our internal approved list. We approve a tool only after a documented security and privacy review.
- Approved use cases only. All AI use sits within an approved use case. Each use case has an accountable use case owner responsible for its safe operation, an impact assessment, a documented risk treatment plan and ongoing review.
- Data classification rules. What data may go into which AI tool depends on its classification. Personal information and OFFICIAL: Sensitive information are only allowed on tools specifically authorised for them.
- Data sanitisation before prompting. Before submitting content to any AI tool, staff remove client names, personal information, credentials and other sensitive material. Where structured data is involved, manual de-identification procedures apply.
- AI governance committee oversight. Our AI governance committee, chaired by our Accountable Official, reviews approved use cases, approves medium- and high-risk use cases, reviews material AI incidents and provides strategic oversight.
- AI incident reporting. We treat AI incidents as a type of security and operational incident. Any event involving AI that causes, or could plausibly cause, harm is reported, investigated and fixed. The same applies to near misses.
- Regular review. We review use cases at least every quarter. We reassess tools every year or on any material change. We review the framework, policies and procedures at least once a year.
These measures apply equally when we build or operate AI inside a client’s systems. We help clients put similar measures in place in their own environments.
How we comply with the DTA Policy
We adopt the DTA Policy v2.0 as our internal standard, even though we’re not subject to it. Our approach follows the policy’s three pillars:
- Strategy and oversight. We have a designated Accountable Official (named under Roles and accountability), keep an internal register of AI use cases and publish this statement. Where we support a non-corporate Commonwealth entity, we cooperate with that entity’s own DTA Policy obligations.
- Preparedness and operations. We operate AI under our internal AI rules and procedures. We provide mandatory training, run incident reporting and remediation, and align how we buy AI products and services with the Australian Government’s published guidance.
- AI use case impact assessment. We screen every AI use case with the National AI Centre’s screening tool and assess it against the DTA Policy’s in-scope criteria. In-scope use cases get a full impact assessment covering ethics, privacy, security and fairness. We use the Australian Government AI impact assessment tool or an equivalent internal process.
We also align with Australia’s AI Ethics Principles, and refer to the ten guardrails of the Voluntary AI Safety Standard.
For the detailed requirement-by-requirement mapping, see Salsa Digital and the DTA AI Policy: how we support client compliance.
Compliance with legislation and regulation
We comply with all applicable Australian legislation and regulation in our use of AI. The most relevant are:
- The Privacy Act 1988 and the Australian Privacy Principles (APPs), including APP 8 on cross-border disclosure of personal information
- The PSPF, including PSPF Policy Advisory 001-2025 on using OFFICIAL information with generative AI
- The Copyright Act 1968, including reasonable steps to avoid accidentally reproducing copyrighted material in AI-generated content
- Client-specific contract requirements, which may set tighter controls than our defaults (where they conflict, the stricter rule applies)
We monitor changes to the legal and regulatory environment for AI, and update our internal documents and this statement in response.
Roles and accountability
Accountable Official. Salsa Digital has designated Paul Morriss as the Accountable Official for AI, as the DTA Policy uses the term. The Accountable Official is accountable overall for putting our AI governance framework in place, maintains our register of AI use cases, approves in-scope AI use cases and is the escalation point for AI incidents.
Accountable use case owners. Each in-scope AI use case has a designated owner with day-to-day responsibility for its safe operation.
AI governance committee. The committee provides strategic oversight of AI use across Salsa Digital. Members include the Accountable Official, security and privacy leads, and senior delivery leads.
In client work, the owner of any AI use case Salsa Digital builds inside a client system sits within Salsa Digital. But accountability for the use case itself sits with the client’s own Accountable Official, under their DTA Policy obligations.
Raising concerns about our use of AI
You can raise concerns about Salsa Digital’s use of AI at any time, including suspected safety issues, unintended behaviour or potential harms. Email security@salsa.digital.
If you’re a member of the public with a concern about AI in a public-facing system we’ve built or operate for a government client, raise it first through that client’s own AI transparency statement. We’ll support our clients in triaging these reports.
We don’t retaliate against anyone who raises a concern or makes a report in good faith under this statement.
Review and maintenance of this statement
We review this statement at least once a year. We update it sooner if there’s a material change to our approach to AI, our tools, our providers, or relevant policy and law. We tell clients about material changes through our normal engagement channels.
If an approved AI vendor materially changes how it handles data, trains models, retains data or where it stores data, and that affects the controls we’ve documented for a client engagement, we notify the affected client within five business days.
We keep the version history internally. It’s available on request.