Amnesty International warned on Thursday the 28th that generative artificial intelligence systems operate, from their own design, with massive collection of personal data without consent. The organization places the problem in models and tools already widespread in the market, including OpenAI's GPT-3, Google's Gemini, Meta's Llama, DeepSeek, Midjourney, and Stable Diffusion.
The tension pointed out by the report is at the core of the business. These tools are presented as products of efficiency and sophistication, but their operation depends, according to the investigation, on illegal extraction of information published on the Internet, including images and social media activity, with effects on privacy, discrimination, and resource consumption.
Amnesty places massive collection of personal data in the design
The investigation, titled Unlawful by Design documents privacy violations, argues that generative AI is trained with billions of public posts and images collected automatically from the web. Amnesty International adds that this practice is carried out without the express consent of the affected individuals.
Likhita Banerji, Director of Accountability on the Use of Algorithms at Amnesty International, stated that companies worldwide market these systems under a guise of efficiency when, in reality, they perpetuate massive privacy violations.
"A different path for technological development is possible if authorities act urgently to correct course" - Likhita Banerji, Director of Accountability on the Use of Algorithms, Amnesty International
The study adds that the training of these models not only affects privacy. It also amplifies content that incites hatred, discrimination, and racial and gender stereotypes, and attributes risks to the right to freedom of thought to generative AI due to its ability to influence ideas and beliefs through predictive suggestions.
Google increased its emissions by 48% and Microsoft increased them by 29%
Amnesty International also links the expansion of generative AI to increased energy consumption, more data centers, and a growing demand for water to sustain this infrastructure. The report connects this deployment with impacts that already appear in the documentation of large technology companies.
In Google's case, its 2024 sustainability report registered that greenhouse gas emissions increased by 48% since 2019, an increase associated with data centers and the supply chain. Microsoft, for its part, increased its emissions by 29% between 2020 and 2024 due to processes supporting AI in its data centers.
Outside of Europe, the organization cites the opposition of communities in Cerrillos, Chile, Querétaro, Mexico, and Arizona, United States, to new facilities in areas affected by droughts and electricity supply problems.
The publication coincided with a call from Pope Leo XIV for regulation
The release of the report coincided with Pope Leo XIV's encyclical Magnifica Humanitas, which warns that whoever controls artificial intelligence will impose their own moral vision and calls on states to set rules.
Amnesty International calls on states to ban autonomous generative AI systems based on illegal extraction of data from the web. It also demands that companies immediately cease the collection of personal data without consent to train models.
The organization further calls for states to hold companies accountable for their involvement in human rights abuses linked to design and business decisions in the development of generative AI.