Skip to main content
Create Free Account

AI & Automation

AI in Supply Chain Compliance: What It Can and Cannot Check

In supply chain compliance, AI reliably reads supplier documents against criteria you set, pulls dates and cover levels off certificates, and finds the outlier in a supplier base too large to read. It cannot verify that a document is genuine, and a supplier's declaration stays a declaration. Australia's Modern Slavery Act 2018 reporting threshold is AU$100 million in annual consolidated revenue, and a statement is due within six months of the end of the reporting period.

  • Published
  • Updated
  • 8 min read
Aerial view of a container port at dusk, with gantry cranes over stacked containers and a network of glowing lines and an AI chip icon drawn over the terminal
Every container here belongs to a supplier somebody had to check. The checking is the part that does not scale.

Supply chain is where the people I build for ask about AI most hopefully, and the reason is arithmetic. A safety team with 40 contractors reads everything that arrives. A procurement team with 900 suppliers does not, and that gap is where supplier risk lives.

This post is about the part of supplier risk that is not safety: modern slavery reporting, insurance and licence currency, sanctions, and documents at a scale nobody covers by reading. Contractor safety prequalification is a different job, and AI contractor prequalification sets out what an agent reviews there and what a person still decides.

Supplier Risk Is Wider Than Safety.

Start with the part that carries a legal duty.

Section 19 of the model Work Health and Safety Act requires a person conducting a business or undertaking to ensure, so far as is reasonably practicable, the health and safety of workers engaged, or caused to be engaged, by the person. Section 46 requires everyone holding a duty in relation to the same matter to consult, co-operate and co-ordinate activities with all the others, so far as is reasonably practicable.1

Safe Work Australia is blunt about a labour hire arrangement: the agency and the host are both persons conducting a business or undertaking, both responsible for the worker’s health and safety, and they cannot contract out of or transfer their duties to each other or to another party.2

Nothing in a supply chain transfers an obligation. It only moves the work of checking, and checking is the thing that does not scale.

Around the safety file sit four questions a procurement team owns rather than inherits: whether a supplier is inside the modern slavery reporting regime, whether their insurance and licences are current today, whether they appear on a sanctions list, and whether anyone has read the documents in it.

Modern Slavery Reporting Starts at AU$100 Million.

This is the fact recalled wrongly most often in supplier compliance, so here it is from the Attorney-General’s Department’s own guidance rather than from memory.

An entity must report under the Modern Slavery Act 2018 (Cth) if it has consolidated revenue of at least AU$100 million over its 12-month reporting period, and is either an Australian entity at any time in that period or a foreign entity carrying on business in Australia at any time in that period. The reporting period means the financial year or other annual accounting period the entity already uses. The statement goes to the Attorney-General’s Department through the online register within six months after the end of that period, and it must address seven mandatory criteria, including the risks in the entity’s operations and supply chains and the actions taken to address them.3

The threshold has not moved. Professor John McMillan AO’s statutory review, tabled on 25 May 2023, made 30 recommendations, one of which was to lower the reporting threshold from AU$100 million to AU$50 million. The government responded on 2 December 2024. As at September 2026 the department’s page still states that the requirement applies at annual consolidated revenue of at least AU$100 million.4 Statements are published on the Modern Slavery Statements Register, which the government must maintain and which anyone can read.5

Two things follow for a supplier program. Suppliers under the threshold owe no statement at all, and asking a small subcontractor for one is how a questionnaire loses its credibility. If you are over the threshold, the answers you collect from suppliers are inputs to your own report.

This is why ComplyFlow’s prequalification form treats modern slavery differently from safety. The section runs ten questions covering prior bribery, corruption or human rights breaches, training, overseas sourcing, labour hire, recruitment fees, written contracts, grievance channels, and cash payment, and the contractor’s responses are not AI-reviewed. They are there to give you visibility into labour practices, risk awareness, and potential areas of concern across your contractor base, and certain answers require a written explanation.6 The supplier-facing guidance says the same from the other direction, telling a small business that a statement is only mandatory once annual consolidated revenue exceeds AU$100 million, or they publish one voluntarily.7

I would not change that design. A model reading a declaration can tell you the declaration is complete. It cannot tell you it is true. For an ESG or sustainability lead the useful output is not a score. It is ten answers in a comparable shape across every supplier, with the ones that required an explanation already surfaced.

Insurance, Licences, and the Things That Expire.

This is narrow extraction work, and AI is genuinely dependable at it: reading a policy number, a cover level, a name, and an expiry date off a scanned certificate.

In ComplyFlow’s prequalification, each insurance question a supplier answers creates a compliance requirement on their profile: the level of public liability cover, workers compensation by state, personal accident, and professional indemnity. They must maintain what they declared and provide current evidence as policies renew, and where a policy expires or evidence is not provided, their compliance status reflects it.6 Behind that sits the requirement set: a bundle of documents, forms, and training you build once, after which ComplyFlow runs the invitations, reminders, and renewals.8 That is what keeps a supplier record current rather than accurate on the day it was collected.

Financial checks are where I stop. Credit assessment and financial distress monitoring come up in every procurement conversation, and I cannot point you at documentation for them, so I will not describe them as something we do. The limit is worth saying out loud: a certificate of currency tells you a policy existed on the day it was issued. It does not tell you the supplier can pay its subcontractors next month.

Sanctions Screening, Where a Match Is Only a Flag.

The Australian Sanctions Office, inside the Department of Foreign Affairs and Trade, maintains the Consolidated List: every individual, entity, and vessel subject to Australian sanctions, including targeted financial sanctions, travel bans, arms embargos, and maritime sanctions. Published as a spreadsheet and updated regularly, it was last updated on 8 September 2026. DFAT describes it as a key resource for due diligence checks.9

The stakes are not procedural. Dealing with a listed individual or entity is a serious criminal offence, with penalties including up to 10 years in prison for individuals and significant fines for individuals and bodies corporate.9

Name matching against that list is work software does better than a person: aliases, transliterations, and near-identical trading names across three entities. A confident false positive costs a supplier a contract; a confident false negative costs you a prosecution. DFAT settles who decides. If you identify a match on the Consolidated List, the instruction is to seek legal advice before proceeding with any dealings involving that person, entity, or vessel.9 A screening tool produces a candidate. A person, usually with a lawyer, produces the answer.

Adverse media screening is the weaker sibling. A model reading news is reading allegations, not findings. Summarising what has been published about a supplier is real work; treating a search result as an adverse finding is not, and it is how a supplier loses a contract over somebody else’s unresolved court case.

Document Review at Supplier Scale.

Two mechanisms, both documented.

The first is the review agent. You build it through a guided wizard: choose the document type, define the review criteria, and keep, edit, or weight the criteria the wizard suggests. Before you save it you run the agent against a real sample document and read the actual result, with up to two free previews that consume no tokens.10

The second is asking your own record a question instead of building a report. ComplyFlow is one of the first compliance platforms with an MCP server, the open standard that lets an assistant such as Claude or GitHub Copilot read another system. Which contractors at a given site have expired insurance documents becomes a question you ask rather than a report somebody builds. Access is read only, so an assistant can look up and summarise but cannot create, change, or delete anything; the token is scoped to the modules you choose, inherits your own permissions and site assignments, and can only reach your organisation’s data.11 The integration side is plumbing. The change is that a question about 900 suppliers no longer needs a report writer and two days.

The Four Things AI Cannot Do Here.

Verify that a document is genuine. It reads what is on the page, and a certificate that was never issued reads exactly like one that was. Where the cover genuinely matters, verification is a call to the insurer or the issuing body.

Turn a declaration into evidence. A supplier’s answer about recruitment fees is a statement about themselves. Collecting it in a comparable format across every supplier is valuable, and it is not an audit.

Carry the duty. Sections 19 and 46 put the obligation on a person, and in labour hire both the host and the agency hold it and neither can transfer it.12 A grade produced by a model is not a defence.

Be trusted without a named human check. Parasuraman and Manzey, reviewing the evidence in Human Factors in June 2010, concluded that automation bias “occurs in both naive and expert participants, cannot be prevented by training or instructions, and can affect decision making in individuals as well as in teams”.12 Safe Work Australia now expects you to treat this as a hazard: a person conducting a business or undertaking must manage health and safety risks from AI and digital technologies as much as they reasonably can, just as with any other hazard, and must consult workers before introducing digital technology that may affect health and safety.13

What This Changes for a Procurement Team.

You do not get a supply chain you can stop checking. You get a shorter list of things to check, on a supplier base you were previously sampling and calling a process.

Pick one thing to do this quarter. Take your supplier list, split it at AU$100 million in consolidated revenue, and stop asking the smaller half for a modern slavery statement they do not owe. Then look at what your reviewers spend their week on, and count how much of it is establishing whether a submission is complete rather than whether a supplier is any good. That second number is the one AI moves. The judgement underneath it was never the bottleneck.

For the safety half of this argument, AI for workplace safety covers what AI is reliable for and what it should never be handed.

Sources

  1. Model Work Health and Safety Bill, sections 19 and 46 Safe Work Australia, Model Bill dated 5 December 2025
  2. Labour hire: WHS duties Safe Work Australia, Read 12 September 2026
  3. Commonwealth Modern Slavery Act 2018: Guidance for Reporting Entities Attorney-General's Department, May 2023 edition
  4. Modern Slavery Act Attorney-General's Department, Read 12 September 2026
  5. About the Modern Slavery Statements Register Attorney-General's Department, Read 12 September 2026
  6. AI Prequalification Form - Standard (Level 1) ComplyFlow Help Centre, Article dated 13 July 2026, read 12 September 2026
  7. Modern Slavery, supplier guidance ComplyFlow Help Centre, Article dated 10 August 2026, read 12 September 2026
  8. Onboarding Requirements ComplyFlow Help Centre, Read 12 September 2026
  9. Consolidated List Australian Sanctions Office, Department of Foreign Affairs and Trade, List last updated 8 September 2026
  10. Release Notes 29-06-2026 AI Tools V2 ComplyFlow Help Centre, Release dated 29 June 2026, article published 6 July 2026
  11. ComplyFlow MCP Server Overview ComplyFlow Help Centre, Article dated 12 June 2026, read 12 September 2026
  12. Complacency and Bias in Human Use of Automation: An Attentional Integration Parasuraman and Manzey, Human Factors 52(3), 381-410; abstract read on the TRID record, National Academies, June 2010
  13. Artificial intelligence (AI) and digital technologies: WHS duties Safe Work Australia, Read 12 September 2026
John McCann

Written by

John McCannHead of Product, ComplyFlow

John has led ComplyFlow’s product since 2021, including its AI document review, its AI agents, and its MCP server. He writes about what AI can and cannot be trusted to do in safety and compliance work, from building it.

Writes about: AI in compliance, Product and integrations, Data and reporting

Questions

Questions People Ask About This.

What is the Modern Slavery Act reporting threshold in Australia?

An entity must report if it has annual consolidated revenue of at least AU$100 million over its 12-month reporting period, and is either an Australian entity or a foreign entity carrying on business in Australia at any time in that period. The reporting period is the financial year or other annual accounting period the entity already uses, and the statement is due within six months of the end of it.

What can AI actually do in supply chain compliance?

Four things well. Read a supplier document against criteria you defined and report what it found. Pull structured detail such as a policy number, a cover level, and an expiry date off a scanned certificate. Match names against a published list at a volume no person would attempt. And answer a question about your own supplier record without somebody building a report first.

Can AI verify that a supplier's insurance certificate is real?

No. It reads what is on the page, and a certificate that was never issued reads exactly like one that was. AI can tell you the cover level, the policy number, and the expiry date, and flag the ones that have lapsed. Where the cover genuinely matters, verification is a call to the insurer or broker, not a document review.

Should AI decide whether a supplier is on a sanctions list?

It should produce the candidate match, not the answer. Name matching across aliases and transliterations is exactly what software is good at, but DFAT's instruction on finding a match on the Consolidated List is to seek legal advice before proceeding with any dealings involving that person, entity, or vessel. The consequence of getting it wrong runs to criminal penalties.

Does a modern slavery questionnaire tell me my supply chain is clean?

It tells you what your suppliers have declared, which is a different and more useful thing than it sounds. Declarations give you a comparable picture across every supplier and show you where the answers required an explanation. They are not audit evidence, and no amount of AI review converts a declaration into one.

See it against your own contractors, sites, and rules.

Book a 30-minute demo. We will show ComplyFlow working with your kind of sites, your kind of contractors, and your requirements. No slides, no hard sell.

  • ISO 27001
  • Hosted on AWS
  • Microsoft & Okta SSO
  • API & MCP
  • Data in Australia

Compliance you can prove, instantly.