Wage theft has been a criminal offence in Australia since 1 January 2025: intentionally underpaying wages or entitlements can now lead to prosecution under the Fair Work Act 2009. Genuine mistakes are not criminal, so your best protection is a consistent, documented payroll process. AI helps by applying award rules uniformly, reconciling paid against owed, and keeping an audit trail.
Key takeaways
- The offence applies to intentional underpayments occurring on or after 1 January 2025, introduced by the Closing Loopholes Act 2023.
- Maximum penalties include up to 10 years' imprisonment for individuals and large fines for companies.
- The Voluntary Wage Compliance Code safe harbour applies only to employers with fewer than 15 employees.
- AI cannot prove or disprove intent. It can make a "genuine mistake" position credible by showing a consistent, monitored process.
- Automating a wrong award mapping industrialises the error. Validate rules with people who know the award before going live.
Is wage theft a criminal offence in Australia?
Yes. On 1 January 2025, intentionally underpaying an employee's wages or entitlements became a criminal offence under the Fair Work Act 2009 (Cth). The change came through the Fair Work Legislation Amendment (Closing Loopholes) Act 2023 and applies to underpayments on or after that date. For two decades an award breach was a civil matter. Where conduct is intentional, the same facts can now put a director or payroll manager before a criminal court.
The Fair Work Ombudsman sets out the maximum penalties:
- Individuals: up to 10 years' imprisonment, a fine, or both. The fine is the greater of three times the underpayment or 5,000 penalty units.
- Companies: a fine of the greater of three times the underpayment or 25,000 penalty units.
At current penalty unit values that runs into the millions for a company and over a million for an individual.
The offence targets intentional conduct. The Fair Work Ombudsman states that accidental underpayment based on a genuine mistake does not attract criminal liability. The uncomfortable question is how you demonstrate, after the fact, that a systematic underpayment was a mistake. Your systems, records and consistent good-faith process become the whole game.
The four things that decide your exposure
- Whether an underpayment happened at all (the debt remains even where no offence is made out).
- Whether the conduct was intentional or a genuine mistake.
- Whether you can evidence a consistent, good-faith compliance process.
- Whether you fall inside or outside the Voluntary Wage Compliance Code.
Why are mid-sized employers more exposed than they think?
There is a comfortable assumption that prosecutions are for deliberate bad actors. Those cases exist. For most employers, though, the structural exposure comes from award complexity meeting manual processes at scale.
Modern awards are intricate: overtime thresholds, casual loadings, split-shift allowances, weekend and public holiday penalty rates, annualised salary reconciliation, task-based allowances, and classifications that turn on duties rather than title. One award can live in a payroll officer's head. Three awards across several states, with full-time, part-time and casual staff, cannot. Manual interpretation at that scale is where systematic errors start.
The key structural fact: the Voluntary Wage Compliance Code, which started alongside the offence on 1 January 2025, gives a safe harbour. If an eligible employer complies with the Code, the Fair Work Ombudsman cannot refer the conduct for criminal prosecution. Eligibility is limited to employers with fewer than 15 employees. Above that, you carry the full offence without the codified pathway out.
The enforcement record points the same way. In its 2024-25 Annual Report, the Fair Work Ombudsman reported recovering $358 million for more than 249,000 underpaid workers, with about 60 per cent coming from large corporate employers. It issued 1,220 Compliance Notices and secured its highest single-action penalty, $15.3 million against the operators of the Sushi Bay outlets. Larger, more complex payrolls are squarely in view.
| Under 15 Employees | 15 to 500 Employees | Impact | |
|---|---|---|---|
| Voluntary Code safe harbour | Available | Not available | Higher exposure |
| Award interpretation load | Often single award | Multiple awards and states | Higher complexity |
| Manual payroll feasibility | Workable | Error-prone at scale | Systemic risk |
| Regulator attention | Lower | Rising with headcount | Greater scrutiny |
| Evidence burden if audited | Smaller dataset | Years of complex records | Harder to reconstruct |
No Code protection, high award complexity and manual interpretation together produce exactly the systematic, hard-to-explain error the offence is designed to catch, whether or not anyone intended it.
Where does AI help with payroll compliance, and where can it not?
Start with the limits. AI cannot form or negate intent and cannot give legal advice. If you feed it the wrong classification, it will underpay every affected employee consistently, which is arguably worse than a person getting it wrong once.
What AI does well is narrower and useful: it applies rules consistently, reconciles large datasets, flags anomalies and keeps a durable audit trail. In a criminal liability context, that audit trail is the artefact that distinguishes a mistake from a decision.
- ClassifyMap each role to the correct award and level
- CaptureRecord hours, breaks, shifts and allowances
- InterpretApply award rules to the recorded data
- ReconcileCompare paid vs owed, flag variances
- EvidenceLog every rule and decision for audit
Classify (a person decides). AI can suggest a likely award and level from duties, and flag when actual tasks drift from the recorded classification. Signing off Level 4 versus Level 3 has legal consequences and stays with a person. Misclassification is the most common root cause of systematic underpayment.
Capture (clean data). Interpretation is only as good as the hours, breaks and allowances feeding it. AI-assisted timekeeping normalises messy inputs, catches missing breaks and reconciles rostered against actual hours. See our guide to automating timesheets.
Interpret (automate). Applying hundreds of rules to thousands of shifts the same way every time is what software should do. A versioned rule engine is far easier to defend than a spreadsheet maintained from memory. More in our piece on Fair Work compliance automation.
Reconcile (automate). Continuous paid-versus-owed comparison turns an annual back-pay scramble into a monthly variance report. See automated compliance reporting.
Evidence (by design). Every classification, rule version, calculation and correction, logged and timestamped, supports a genuine mistake position. Designing it with access controls and human override points is covered in our AI agent governance framework.
| Manual Process | AI-Assisted | Impact | |
|---|---|---|---|
| Award rule application | Inconsistent, memory-based | Versioned, applied uniformly | Consistency |
| Error detection | Annual or on complaint | Continuous variance flags | Earlier |
| Audit trail | Fragmented spreadsheets | Timestamped decision log | Defensible |
| Back-pay exposure | Compounds over years | Caught and fixed monthly | Contained |
| Reviewer time per cycle | High, reactive | Lower, exception-based | Focused effort |
How does a documented process help prove a genuine mistake?
No software can prove a state of mind. It can make the innocent explanation credible. For example, take a services employer that finds, during a routine review, that staff on annualised salaries were never reconciled against hours worked and were underpaid over 18 months. The debt must be back-paid in full. The criminal question is intent.
Two versions of that employer fare very differently:
- Version one ran payroll on spreadsheets, had no documented reconciliation, and cannot show when or how the error entered.
- Version two has a versioned rule engine, a monthly reconciliation report showing the variance was flagged and being actioned, a log of every classification decision, and a disclosure to the regulator as soon as the issue was confirmed.
Same debt, very different position on intent. Version two looks like what it is: an employer with systems designed to catch and correct mistakes.
Self-reporting matters too. The Fair Work Ombudsman has indicated that a voluntary, frank and complete disclosure may allow an employer to enter a cooperation agreement, under which the regulator agrees not to refer the conduct for prosecution. Systems that surface problems early let you self-report early.
What is the nature of the task?
- Applying award rules to recorded hoursAutomate: consistency is the goal
- Reconciling paid vs owed across cyclesAutomate: volume and repetition
- Assigning an award classificationHuman decides, AI advises
- Deciding to self-report to the regulatorHuman and legal counsel only
A caution: automating a wrong rule set industrialises the error. Before any award logic goes live, someone who understands the award, ideally with employment law input, must validate the mapping. That validation is never delegated to the tool that will run it.
What are the common underpayment traps?
Most large underpayment findings come from a handful of structural traps where complexity and manual process repeat the same error across hundreds of pay cycles. Point your audit and automation here first.
- Annualised salary reconciliation. Many awards that permit annualised salaries require reconciliation against what the employee would have earned for hours actually worked, with any shortfall topped up. A salaried employee who regularly works long hours can accrue an underpayment while their pay never changes.
- Casual and part-time treatment. Missed casual loadings, minimum engagement periods, casual conversion rules, and additional part-time hours that should attract overtime.
- Task-based allowances. Tool, travel, first aid, height or dirty-work allowances that turn on what someone did that day. The trigger sits in the timesheet, and manual payroll often misses it.
- Penalty rates and overtime thresholds. Weekends, public holidays and shift patterns interact in ways spreadsheets handle badly.
| Where It Hides | What Catches It | Impact | |
|---|---|---|---|
| Annualised salaries | No reconciliation of salary vs hours worked | Continuous paid-vs-owed comparison | Early |
| Casual and part-time | Flat rates, missed loadings and overtime | Rule-based entitlement checks | Consistent |
| Task-based allowances | Trigger sits in the timesheet, unnoticed | Allowance flags from captured job data | Complete |
| Penalty and overtime | Interacting rules break spreadsheets | Versioned rule engine | Reliable |
None of these need bad intent to create a large liability. Automating reconciliation and rule application across these four populations gives the most risk reduction for the least effort.
What does a realistic 90-day payroll compliance uplift look like?
You do not fix years of payroll practice in a fortnight. Prioritise the highest-risk awards and classifications, validate before you automate, and build the audit trail from day one.
- Weeks 1-3Audit and mapReview classifications and award mappings for the highest-risk populations, with employment law input
- Weeks 4-7Validate rulesEncode and test award logic against known-correct cases before any live use
- Weeks 8-10ReconcileRun paid-vs-owed reconciliation over recent cycles, surface and cost any variances
- Weeks 11-13Monitor and evidenceMove to continuous variance monitoring with a full audit log and a self-report protocol
The business case differs from most automation cases. The biggest line item is avoided liability, which is real even when it is hard to put one number on. Saved hours are worth quantifying for your environment, but the compliance case is what gets board sign-off.
- Reviewer hours redirected from manual checking to exception handlingOngoing
- Back-pay caught monthly instead of compounding over yearsContained
- Documented good-faith process supporting a 'genuine mistake' positionDefensible
- Early detection enabling voluntary self-report and cooperationOptionality
- Board-level assurance that award risk is actively monitoredGovernance
Two cautions. First, this does not remove human accountability, and it should not be sold as a payroll headcount cut. The people who know your awards move from manual calculation to validation and exception handling, which is a change management task covered in our guide to AI adoption in Australian workplaces. Second, it is no substitute for legal advice. Whether a past underpayment carries criminal exposure is a question for a qualified employment lawyer.
How Solve8 can help
Our consultants and engineers design award interpretation, reconciliation and audit trail processes that run on your infrastructure in Australia, with a person approving classifications and corrections. See our process automation and system integration services, or talk to us about your award exposure.
