Voluntary AI Safety Standard
The guardrails are voluntary. They do not create a new legal duty for a law firm.
Ten voluntary guardrails for organisations that develop or deploy AI in Australia. Later guidance evolves them. They do not create new legal duties.
MaintainedLast reviewed
Key facts
- Issued by
- Department of Industry, Science and Resources
- Sources listed
- 3
This is a reading aid for an Australian law firm. It is not legal advice, not a certification, and not a finding that a tool or a firm complies. The sentences paraphrase the sources listed, checked on the date shown. Where a standard is sold by ISO or the AICPA, the page uses the publisher’s public description and does not reproduce the standard.
What it is
The Department of Industry, Science and Resources published the Voluntary AI Safety Standard to give practical guidance to Australian organisations on using and innovating with AI safely. The department says the standard consists of 10 voluntary guardrails that apply to organisations throughout the AI supply chain, including transparency and accountability, and that they explain what developers and deployers of AI systems do to meet the guardrails.
The guardrails page states that on 21 October 2025 the department published Guidance for AI Adoption, which outlines 6 essential practices for safe and responsible AI governance, and that this updated guidance evolves the Voluntary AI Safety Standard. This page describes the 10 guardrails from the department’s guardrails page. It does not restate the later 6 practices. Read that guidance before treating the 10 as the department’s last word.
The OAIC says the standard does not seek to create new legal obligations. It is meant to help organisations use AI in accordance with laws that already apply.
Who it applies to
The department says the guardrails apply to all organisations throughout the AI supply chain. “Apply” here means the standard is written for them. It is voluntary. A law firm that buys or uses an AI system is a deployer in the sense the standard uses. Nothing on the page makes the guardrails a condition of practising law.
What it asks of a law firm using AI
- Guardrail 1: establish, implement and publish an accountability process, including governance, internal capability, and a strategy for regulatory compliance.
- Guardrail 2: establish and implement a risk management process to identify and mitigate risks.
- Guardrail 3: protect AI systems, and implement data governance measures to manage data quality and provenance.
- Guardrail 4: test AI models and systems to evaluate model performance, and monitor the system once it is deployed.
- Guardrail 5: enable human control or intervention in an AI system to achieve meaningful human oversight.
- Guardrail 6: inform end-users about AI-enabled decisions, interactions with AI, and AI-generated content.
- Guardrail 7: establish processes for people impacted by AI systems to challenge the use or the outcome.
- Guardrail 8: be transparent with other organisations across the supply chain about data, models and systems, so they can address risks.
- Guardrail 9: keep records to allow third parties to assess compliance with the guardrails.
- Guardrail 10: engage stakeholders and evaluate their needs and circumstances, with a focus on safety, diversity, inclusion and fairness.
- The department says that to adopt the standard completely, an organisation adopts all 10, and that a practical start is guardrail 1.
How it maps to this framework
The rows are this site’s questions. Where the cell says “Not set here”, the instrument does not set that requirement. The gap is still recorded, so it is not treated as answered.
| This site’s question | What the instrument says that bears on it |
|---|---|
| Efficiency Whether the tool saves total human work at the quality the task needs. | Not set here. |
| Effectiveness Whether the output is good enough for the task. The catalogue does not score this. | Guardrail 4 is testing and monitoring performance after deployment. That is about the system behaving as intended. It is not a substitute for the lawyer reading the output. |
| Due diligence What the firm has actually read about hosting, training and assurance before it relies on the tool. | Guardrails 1, 2, 3, 8 and 9 are the governance, risk, data, supply-chain and record questions a firm can ask a vendor and ask of itself. Voluntary does not mean the answers are optional inside this framework. It means the department has not made them a legal duty. |
| Confidentiality Whether client and other confidential information can go into the tool. | Guardrail 3 is about protecting the system and governing data quality and provenance. It does not restate a solicitor’s duty, and it does not authorise a particular upload. |
| Disclosure When the firm tells the client, and where required the court or another party, that AI was used. | Guardrail 6 is about informing end-users when they are dealing with an AI decision, an AI interaction, or AI-generated content. Guardrail 7 is about a way to challenge the use or the outcome. Those are the standard’s transparency ideas. They are not a court practice note, and they do not say every client letter must name the tool. |
| Certifications ISO, SOC 2 and IRAP names recorded against the tool. A listed name is not a finding that the certificate covers the firm’s plan. | The standard does not say that ISO/IEC 42001, or any other certificate, satisfies the guardrails. A certificate can be one record under guardrail 9. It is not the guardrail. |
| Data residency Where the tool holds data, including whether that place is recorded as Australia. | Not set here. |
Sources
- Voluntary AI Safety Standard
- Voluntary AI Safety Standard: the 10 guardrails
- Voluntary AI Safety Standard: introduction
Checked 2026-10-10. Issued by Department of Industry, Science and Resources.