Forced Labour and Human Rights Due Diligence
Deliberate Academy Editorial Team
Reviewed for accuracy and professional relevance
You're 6 lessons in — don't lose your progress.
Sign up free to save where you are and earn a verified certificate when you pass.
- Distinguish the obligations imposed by the EU CSDDD, the UK Modern Slavery Act, and the US UFLPA, and identify which apply to your organisation
- Explain why the UFLPA rebuttable presumption reverses the evidential burden and what that requires of your records
- Use AI for screening and document processing without letting it substitute for the worker-level evidence these regimes ultimately require
- Build a due diligence record that evidences reasonable steps rather than a clean result
This is the lesson where sourcing becomes a legal exposure rather than a commercial one. Three regimes matter most, they impose different obligations, and organisations frequently apply the logic of one to a situation governed by another.
Three Regimes, Three Different Asks
UK Modern Slavery Act 2015 (section 54). A transparency regime. Commercial organisations above a turnover threshold carrying on business in the UK must publish an annual slavery and human trafficking statement, approved by the board and signed by a director. It does not mandate specific due diligence. What it does mandate is that you state what you do, which creates a real exposure: a statement committing you to supply chain due diligence that you cannot evidence is a problem of your own making, and the statements are public and compared.
EU Corporate Sustainability Due Diligence Directive (CSDDD). A conduct regime, and a substantially higher bar. In-scope companies must identify actual and potential adverse human rights and environmental impacts across their chain of activities, prevent or mitigate them, provide remediation where they caused or contributed to harm, and operate a complaints mechanism. It requires action, not disclosure, and it extends beyond direct suppliers. Member state transposition governs the detail, so confirm the position in the states where you operate rather than working from the directive alone.
US Uyghur Forced Labor Prevention Act (UFLPA). An import control regime with a mechanism unlike the other two. Goods wholly or partly produced in the Xinjiang Uyghur Autonomous Region, or by entities on the UFLPA Entity List, are subject to a rebuttable presumption that they were made with forced labour and are barred from entry to the United States. Customs and Border Protection detains, and the importer must rebut with clear and convincing evidence.
That reversal is the practical heart of this lesson. Under the other regimes you must show you took reasonable steps. Under UFLPA, once the presumption attaches, you must prove a negative about a specific shipment's origin, to a demanding standard, while the goods sit detained.
The UFLPA rebuttable presumption means your evidence must exist before the detention, not be assembled after it. Rebuttal requires complete, verifiable supply chain tracing for the specific goods — purchase orders, production records, transportation documents, and worker-level evidence back to raw material. A supplier attestation that they do not use forced labour is not sufficient and will not rebut.
What UFLPA Rebuttal Actually Requires
Because the standard is unfamiliar, it is worth stating what a rebuttal package contains. CBP guidance expects tracing of the specific detained goods through every production stage back to raw material, evidenced by records: purchase orders and invoices at each tier, production and processing records, transportation documents linking each stage, and evidence about the workforce at each facility.
This is documentation that has to be collected in the ordinary course of business. It cannot be reconstructed after detention, because the records sit with tier-two and tier-three parties who have no contractual relationship with you and no reason to respond quickly to a company whose goods are already stopped.
The implication for sourcing is direct. For any input with plausible exposure, the traceability documentation requirement belongs in the contract before the first order, with the flow-down structure from lesson five. A company that identifies its exposure only when a container is detained has already lost the argument, whatever the truth of the matter.
Where AI Genuinely Helps
Within these regimes AI does substantial useful work, all of it upstream of the judgment.
Entity screening at scale. Matching your supplier base and disclosed tier-two entities against the UFLPA Entity List and other restricted lists, continuously rather than at onboarding. Names transliterate inconsistently and entities restructure, so fuzzy matching is genuinely necessary here.
Adverse media monitoring. Searching non-English local media, NGO reports, and litigation records at a scale no team can cover manually. This is where AI's language coverage is a real advantage, and where translation quality matters.
Document processing. Extracting structured data from certificates of origin, audit reports, and supplier disclosures arriving in inconsistent formats and multiple languages.
Risk prioritisation. Combining commodity, region, and process signals to rank where to concentrate limited audit budget.
Detecting patterns across audits. Identifying that several facilities audited by the same firm produce implausibly similar reports, or that a facility's reported working hours are inconsistent with its reported output.
That last one points at the boundary.
What AI Cannot Establish
Forced labour is a condition of employment, not a data pattern. Its indicators are worker-level: retained identity documents, recruitment fees creating debt bondage, restricted freedom of movement, withheld wages, threats.
None of that appears in trade data, corporate records, or a supplier's own attestation. It is established by worker interviews conducted independently and off-site, by unannounced facility visits, by grievance mechanisms workers actually trust, and by credible local civil society reporting. AI can tell you where to send an auditor. It cannot tell you what the auditor would find.
Two consequences follow that are frequently missed.
A clean screening result is weak evidence of absence. Screening establishes that a supplier is not on a list and has not been the subject of reported allegations. Both are properties of what has been documented publicly, and forced labour is characteristically undocumented. A clean result means you found nothing, and in a high-risk category that is a reason for worker-level verification rather than a substitute for it.
Social audits have well-documented limitations. Announced audits, document review conducted with management present, and worker interviews on site are all subject to coaching and retaliation. Audit reports are useful inputs; treating a passed audit as proof is the failure mode that has recurred in most major supply chain scandals of the last two decades.
A US importer's goods are detained under UFLPA. The importer holds signed supplier attestations confirming no forced labour, a passed third-party social audit of the tier-one facility, and a clean screening result against the Entity List. Is this sufficient to rebut the presumption?
Select one answer.
Twelve weeks detained, and the tracing records that did not exist
Context
An apparel company sourced finished garments from a tier-one manufacturer in a third country. The manufacturer was audited annually, held a current social compliance certification, and had provided written confirmation that it did not source cotton from Xinjiang. A shipment was detained by CBP under UFLPA on the basis of isotopic testing indicating the cotton's likely origin.
Action
The company attempted to assemble a rebuttal. The tier-one manufacturer identified its fabric mill. The fabric mill identified its yarn spinner. The yarn spinner was unable or unwilling to identify the specific cotton gin and bale lots for the production run in question, and had no contractual obligation to the apparel company. No records linked the specific detained garments to specific raw cotton, because nothing in the supply chain had ever been required to maintain that linkage.
Outcome
The shipment was excluded after twelve weeks and the goods were sold outside the United States at a substantial loss. The company subsequently rebuilt its cotton sourcing around traceability as a contractual precondition: gin-level origin documentation and bale lot linkage required from the yarn stage onward, flowed down through each tier, with a verification right and a stated consequence for non-provision. The sourcing director's summary was that every compliance instrument the company held had been designed to demonstrate good practice, and none of them had been designed to trace a specific garment to a specific bale, which was the only thing the regime actually asked for.
The UK Modern Slavery Act mandates no particular due diligence activity. Where does this lesson locate the exposure it nonetheless creates?
Select one answer.
Exercise
Your Task
Establish which of the three regimes apply to your organisation, by turnover, EU presence, and whether you import into the United States. For one high-risk category, then do three things. First, identify the raw material stage where the risk originates and count how many tiers separate it from your tier-one supplier. Second, determine whether you could today produce tracing records linking a specific shipment back to that raw material stage, and identify at which tier the chain of records breaks. Third, list the compliance instruments you currently hold for that category — attestations, audits, certifications, screening results — and mark each according to whether it would contribute to a UFLPA rebuttal or only demonstrate general good practice.
Success looks like
- Regime applicability is determined against specific thresholds and jurisdictional tests rather than assumed
- The tier at which the tracing record chain breaks is identified specifically
- Existing instruments are honestly classified, recognising that most demonstrate good practice rather than trace goods
- Where the organisation publishes a Modern Slavery statement, its commitments are checked against what can actually be evidenced
Watch out for
- Treating a clean screening result in a high-risk category as evidence of absence rather than as a prompt for worker-level verification
- Assuming a passed social audit at tier one addresses risk originating three tiers upstream
- The three regimes ask different things: the UK Modern Slavery Act requires a published statement and creates exposure through commitments you cannot evidence, CSDDD requires identification, prevention and remediation across the chain of activities, and UFLPA is an import control operating through a rebuttable presumption.
- UFLPA reverses the evidential burden. Rebuttal requires complete verifiable tracing of the specific goods back to raw material with records at every tier, which must be secured contractually before ordering because it cannot be reconstructed once goods are detained.
- AI does the upstream work well: continuous entity screening with fuzzy matching, non-English adverse media monitoring, document extraction, risk prioritisation, and detecting implausible patterns across audit reports.
- Forced labour is a condition of employment with worker-level indicators — retained documents, recruitment debt, restricted movement, withheld wages — that appear in no dataset. AI can direct an auditor; it cannot establish what the auditor would find.
- A clean screening result means nothing was documented publicly, and forced labour is characteristically undocumented. In a high-risk category that is a reason for independent worker-level verification, not a substitute for it.