The AI Act: when 21 out of 24 documents forget to name the harvested sites, deafening silence!

The main fact
In a striking display of bureaucratic ambiguity, 21 of the 24 AI model training summaries examined failed to name any harvested websites, making it impossible for news publishers to verify whether their content had been used. This situation raises serious questions about the transparency and accountability of AI providers.
What happened
The European Commission implemented the AI Act, requiring providers of general-purpose AI models to publish a detailed summary of their training content. However, of the 24 summaries analyzed, 21 merely mention vague categories of sources without ever citing specific domain names. DeepSeek’s three summaries, meanwhile, completely omit this section. Publishers seeking answers are therefore left entirely in the dark.
The context
This transparency requirement, which came into effect on August 2, 2025 for new models and on August 2, 2027 for those already on the market, aims to allow rights holders to verify the use of their content. The Commission had promised that this list of domains would help identify training sources. But now, as the results are being assessed, it seems that reality falls far short of these promises.
Key figures
Of the 24 summaries reviewed, 21 provide no identifiable information, merely using general terms like « web platforms » or « hosting services. » The three DeepSeek documents don’t even include a section on harvesting. This raises questions about the providers’ compliance with Section 53(d) of the AI Act, which requires clear and specific information.
What this reveals
This situation highlights a double standard: while the EU champions transparency, AI providers seem to be playing the opacity card. Software publishers, supposedly protected by this legislation, find themselves powerless. The promise of a list of domains was meant to free them from the need to request information from providers. On the contrary, it seems to trap them in a bureaucratic maze. An irony that will ly amuse those familiar with how the EU operates: promising transparency only to deliver a thick fog.
Key takeaways
In short, the AI Act, intended to guarantee transparency and accountability, seems to have been a damp squib. AI providers continue to play with words, leaving developers in the dark. This situation calls for urgent reflection on the need for genuine transparency in the AI sector.
Source: ActuIA
