Workplace AI guide: what the federal draft asks of employers

The federal government's draft Best practice guide on the adoption of AI in the workplace asks employers to bring AI in with their staff: involve them early, agree the rules together, train them and keep people accountable.

The minister released the guide on 12 November 2026, and it's open for public comment until 27 January 2027. It's voluntary, and it's still a draft, so its wording may change before the final version. This article sets out what the draft asks of employers, then works through one everyday case: bringing in an AI tool that staff use to write emails and other messages.

Key points

• The guide is voluntary. It creates no new legal obligations, and existing duties under workplace, privacy and work health and safety laws still apply.
• It's built on three themes, Plan, Enable and Embed, and eight principles.
• It asks employers to treat AI adoption as a change to how people work, planned with the people doing the work.
• It asks for AI guidance written with staff, training that includes checking what AI produces, clear human accountability, and privacy and security considered from the moment a tool is chosen.
• Consultation closes on 27 January 2027.

What is the Best practice guide on the adoption of AI in the workplace?

The Best practice guide on the adoption of AI in the workplace is a draft federal guide, released by the Minister for Employment and Workplace Relations, Amanda Rishworth, on 12 November 2026. The minister's media release says the draft came out of roundtables and forums with unions, employer peak bodies, academics, and small and large employers.

The draft guide describes itself as voluntary. It doesn't create or expand legal obligations, it's not a code of practice under work health and safety laws, and it applies in addition to an employer's existing duties under workplace relations, privacy, anti-discrimination and safety law. It's written for workplaces of all sizes, with notes on how small businesses can apply several of the principles. The guide is © Commonwealth of Australia 2026 and published under a Creative Commons Attribution 4.0 International licence; this article draws on it throughout.

The guide's starting point is that AI adoption should be handled as a change to how people work. In its words, best practice treats high-impact AI "as a workplace transformation, rather than a technology rollout". The minister put it this way in the release: "AI works best when the workforce is at the centre of the adoption process."

The three themes and eight principles

The guide groups its eight principles under three themes. It also names five values agreed by the AI Employment and Workplaces Forum of unions, employers and government: trust, capability, transparency, safety and productivity.

Plan: preparing for adoption

1. Capturing the opportunity: set clear goals for how AI can benefit the organisation and its workers.
2. Identifying use cases and assessing workplace readiness: start from a problem the business has.
3. Cooperative planning and worker engagement: involve workers throughout.

Enable: job design and capability

4. Designing work and jobs: redesign tasks around what AI does well.
5. Building capability: give everyone the skills to use AI well.
6. Managing workforce transformation: plan for how roles change, and support the people in them.

Embed: governance, risk management and safety

7. Establishing AI governance: clear structures, accountability and legal compliance.
8. Proactively managing risks and work health and safety.

What the draft guide asks employers to do

The Best practice guide on the adoption of AI in the workplace sets out what good practice looks like under each principle. This article picks out six that bear on everyday AI tools.

Involve staff before the decision is made

The guide asks employers to engage workers early, before a decision to introduce AI is made, and to keep engaging them as the tool is used. It describes three stages: co-design (are we solving the right problem?), engagement in decision-making (are we making the right decision?) and collaborative implementation (is it working as intended?). The engagement should be two-way and in proportion to the size of the change.

The guide also notes that existing consultation obligations under modern awards, enterprise agreements and work health and safety laws still apply when an employer proposes to introduce AI that affects workers. One case study, the Australian Retirement Trust, shows an enterprise agreement with the Finance Sector Union that gives workers a voice where AI is likely to automate roles or be used to judge work performance.

Write the AI guidance with staff

The guide asks employers to give staff clear, accessible guidance on AI use, through existing policies or a dedicated AI policy, written together with workers and, where they have them, their representatives. In Deloitte's State of AI in the Enterprise Australian results, released on 9 November 2026 from a survey of 3,500 leaders across 25 countries, 75% of Australian organisations said they provide employees with AI training and guidelines. Deloitte's release doesn't give the Australian sample size.

Teach people to check what AI produces

The guide lists six areas of AI capability. The first, core AI literacy, includes the ability to judge information produced by AI and to know when it needs extra review, checking or human oversight. That's a practical skill for any tool that produces text a person then sends under their own name. The guide also says capability building works best when it happens during working hours.

Keep people accountable

The guide asks employers to keep accountability for AI "with the decision-makers, not the technology", with an accountable leader for each use of AI. Where AI supports a decision, someone in the business should be identified as responsible for the final call.

Consider privacy and security when choosing the tool

Among the trust factors in the guide, safety includes protecting personal and workplace information, with privacy and security considered from the design or selection of a tool through to its use and review. Ramsay Health Care, one of the guide's case studies, set up an AI governance committee, keeps a central register of AI systems, and runs privacy and cyber assessments before any system is piloted.

Watch workload after the rollout

The guide asks employers to keep monitoring AI after it's introduced, with workers, and to review job satisfaction and workload. Among the work health and safety risks it names is work intensification, where routine tasks are automated and people are left with a heavier share of complex work.

Applying the guide to AI writing and email tools

The Best practice guide does note that AI is often built into existing workplace software, including productivity software, and one of its case studies reports a result in communications. Though not directly mentioned in the guide, email and writing tools fit that category.

In the draft guide (November 2026), Brisbane Catholic Education, which runs 146 schools in Southeast Queensland, reports that an independent evaluation found 70.9% of its educators saved significant time on administrative tasks, "particularly in communications, planning and documentation". The minister's media release rounds the figure to 71%. The guide doesn't give the evaluation's sample size or date, doesn't say which tools staff used, and doesn't separate the time saved on communications from the time saved on planning and documentation. The result also comes from a single education employer.

We read it as relevant to AI writing or email tools because communications are where those tools are used, and the case study names communications as one place staff saved time. On that reading, here's how the guide's principles apply to bringing in an AI tool for writing emails and messages:

1. Ask staff where writing costs them time, and which messages they find hardest to get right, before choosing a tool.
2. Agree with staff which messages the tool should and shouldn't be used for, and write that into your AI guidance.
3. Train people to read the tool's output before they send it, and to know when a message needs a second reader.
4. Make clear that the sender remains responsible for every message that goes out under their name.
5. Ask the vendor where drafts are processed, what personal information is removed or kept, and whether drafts are used to train models, before you sign.
6. After rollout, check with staff whether the tool saves time or adds work, and adjust.

Where Harmony fits

Harmony by Unanimas works in Gmail and Outlook: helping teams understand how their drafts land with their reader, before the message is sent. It offers rewrites in the sender's voice, adapted for the person receiving it. It runs as a sidebar inside the email client staff already use.

On the privacy questions the guide raises, Unanimas states that "PII and PCI data is sanitised before it reaches our translation engine", a commitment it calls Data Serenity™, and that "Harmony's translation engine runs only on Australian servers." In a LinkedIn post on 10 November 2026, Unanimas's founder, Dorian Detragiache, wrote: "We're wrapping up our Stage 1 audit for ISO27001:2022, and kicking off our SOC2 Type 1 audit next week." How those fit an organisation's own requirements is for each employer to judge; visit the Unanimas Trust page for further details.

How to have your say on the draft guide

Submissions on the draft Best practice guide on the adoption of AI in the workplace go through an online form on the department's consultation page until 11:59 pm AEDT on 27 January 2027, and the department may publish them. If your organisation already uses AI tools, including ones staff write with, what has and hasn't worked is useful input.

FAQs: Workplace AI guide: what the federal draft asks of employers

Is the federal workplace AI best practice guide mandatory?

No. The Best practice guide on the adoption of AI in the workplace is voluntary and doesn't create or expand legal obligations. Employers still have to meet their existing duties under the Fair Work Act, work health and safety laws, privacy law and other laws when they introduce AI.

When does consultation on the draft workplace AI guide close?

Consultation closes on 27 January 2027. Comments go through the Department of Employment and Workplace Relations consultation hub, and the guide may change before the final version is published.

Do employers have to consult staff before introducing AI tools?

Sometimes, under existing law. The guide notes that consultation obligations in modern awards, enterprise agreements and work health and safety laws apply when an employer proposes to introduce AI that affects workers. The guide itself asks for more than the minimum: engagement early, before the decision is made, and throughout the tool's use.

Does the guide cover AI email and writing tools?

The guide doesn't name email or writing tools, but its principles apply to AI built into everyday workplace software, which it says includes productivity software. Its points on staff involvement, written guidance, training, accountability and privacy can be applied to a tool staff use to write messages.

What should a workplace AI policy include under the guide?

The guide asks for clear, accessible guidance on how AI is used, written with workers and their representatives where applicable. Based on its principles, a policy for an AI writing tool would cover which messages the tool is for, who is accountable for what is sent, how staff check the output, and how personal information is handled.

Who developed the draft Best practice guide on the adoption of AI in the workplace?

The federal government developed it through roundtables and forums with unions, employer peak bodies, academics and small and large employers, according to the minister's media release of 12 November 2026. Its five values come from the AI Employment and Workplaces Forum, which brings together unions, employers and government.

Where does Harmony by Unanimas process the drafts it translates?

In Australia, according to the Unanimas Trust page, which states that Harmony's translation engine runs only on Australian servers, and that personal identifiers and payment details are sanitised before a draft reaches it. Unanimas publishes its current SOC 2 and ISO 27001 status in its Trust Centre.

To see how Harmony reads a draft for the person receiving it, try the live demo on the Unanimas homepage.