How Amazon’s AI Data Intelligence Detects Labour Violations

Amazon is applying AI to enhance the identification of human rights risks, such as forced labour within its supplier network.

Amazon is using ML and other technologies in a joint effort with the World Economic Forum (WEF) to address human rights issues throughout its worldwide operations.

Amazon has developed and deployed AI models that are designed to detect risks of forced labour among its suppliers.

This system works by analysing a large volume of data points drawn from sources including historical audits, government reports and news signals to flag supplier sites that could be considered high-risk.

Kara Hurst, Amazon's Chief Sustainability Officer, says that the tool "successfully identified about nine out of every 10 high-risk sites with 85% overall accuracy".

This capability allows Amazon to better direct its due diligence resources across its varied business areas, which include e-commerce logistics, cloud computing and manufacturing.

In addition to this risk detection tool, Amazon has also created an AI system to speed up the review of audit reports.

Kara explains that early versions helped process audit reports "65% faster – a remarkable difference" when compared to the typical four hours needed for a manual review.

Integrating human rights principles

Amazon established its formal commitment to human rights in 2019 based on the UN Guiding Principles on Business and Human Rights.

This followed Amazon's first supplier code of conduct, which was published in 2014.

Leigh Anne DeWine, Amazon's Director of Human Rights & Social Impact, told Devex that progress has been achieved by putting these principles into practice through work with individual business units.

She said Amazon transitioned from broad human rights assessments to creating specific due diligence processes for the unique risk profiles of different sectors.

Amazon has also increased its supplier transparency by mapping its suppliers publicly on the Open Supply Hub.

According to Amazon, it addressed all 826 complaints made through its human rights and environmental complaints form last year.

Addressing data fragmentation challenges

A primary obstacle for businesses working to improve human rights protections is data fragmentation. Leigh Anne sees this as a persistent issue.

Human rights problems can often go undetected because information is "incomplete, inconsistent or siloed across owners' countries and systems," she explains. Without common data standards and reliable methods for collaboration efforts, managing risk can become isolated and purely reactive.

A company might identify a problem in a single facility but cannot work with others or address root causes further up the supply chain.

This issue led Amazon to become part of the WEF's Global Data Partnership Against Forced Labour, an initiative aimed at promoting the responsible sharing of data between different sectors.

The limitations of AI in human rights

Despite the potential of AI for risk detection, Leigh Anne emphasised that technology should be seen as a support for human judgement, not a replacement for it.

She believes that "even the most advanced models are only as strong as the data behind them".

Key challenges include improving the quality and scope of data used to train these models and promoting the responsible use of AI across the industry.

To this end, Amazon is working with NGOs, research bodies and audit firms to enhance its data coverage and thoroughly test its AI models.

Amazon's strategic priorities for human rights involve strengthening its policy foundations, integrating these principles more deeply into sectors like logistics and construction and advancing its AI-driven risk analysis.

A central part of this strategy is ensuring that workers' voices are heard.

As Amazon proceeds with its plans to decarbonise its operations, it also intends to manage a just transition by incorporating social factors into its environmental strategies.​​​​​​