Amnesty International calls for banning data-scraping generative models because unlawful training pathways threaten system engineering
The world’s leading generative AI systems rely on automated scraping pipelines that collect billions of posts and personal images without explicit consent from their owners, a practice Amnesty International described as making these systemsinherently unlawful by designA new brief report from the organization stated that tech companies extract massive amounts of publicly available online data using unlawful techniques to develop their products, resulting in widespread violations of users’ privacy.
The organization’s review covered the models powering the most prominent publicly available tools, led by OpenAI’s GPT-3, Google’s Gemini, Meta’s Llama, as well as DeepSeek models and the image-generation tools Midjourney and Stable Diffusion. The report confirmed that training these models was primarily based on harvesting billions of posts and images from social platforms and public sites without explicit permission from their creators or subjects.
Lequita Banerjee, head of the Algorithmic Accountability Lab at Amnesty International, explained that companies worldwide market generative AI products as advances in efficiency and progress, while in practice they institute systematic privacy violations. Banerjee noted that this automated process pulls personal activity and private images and feeds them directly into model weights, turning individual data into free fuel for training commercial algorithms.
The repercussions of this approach do not stop at privacy breaches; the report revealed that the expansion of randomly harvested massive data sets amplifies discriminatory and hate-based content outputs, reproducing negative stereotypes and biases, especially those based on race or gender. Based on these findings, Amnesty International issued a direct appeal to governments worldwide to imposean explicit ban on stand-alone generative AI systemsthat were built using unlawful techniques to scrape and aggregate data at massive scale.
The system’s costs also extend to the environment; Google’s 2024 sustainability report disclosed a sharp 48 % rise in greenhouse-gas emissions since 2019, which the company attributed to data-center demands and supply-chain requirements. In the same vein, Microsoft’s emissions jumped 29 % between 2020 and 2024, driven by expanding data centers dedicated to supporting AI workloads and complex computations.
This global shift creates a new professional reality for data engineers and intelligent-model builders in the Gulf, Egypt and the Levant; random data collection is no longer merely a technical risk but now threatens the viability of entire corporate projects. Financial institutions, government agencies and local development teams are compelled to rigorously verify training-set sources and ensure they are free of legal infringements and discriminatory biases, shifting the regional job focus from merely fine-tuning ready-made models togovernance of data quality and auditing its supply chainsand engineering them in an environmentally and legally responsible manner.
Building sustainable intelligent systems in our markets requires data engineers and software developers to address source auditing and privacy protection as core technical pillars; relying on models tainted by rights violations or unlicensed data jeopardizes the entire product, and true efficiency lies in the ability to sift trustworthy data and manage it professionally.