Artificial intelligence platforms including ChatGPT, Claude and Gemini must embed invisible tracking markers in online text under new European Union transparency regulations.
The new rules will come into application on December 2, requiring AI developers to leave a digital trace on any text generated across the internet.
Technology journalist Anicet Mbida detailed the incoming regulations during an appearance on the French morning broadcast Bonjour La Matinale TF1 on Wednesday, October 7.
Mbida explained that identifying AI-generated material will soon become significantly easier, comparing the tracking process to checking the watermark embedded inside paper banknotes.
According to Mbida, determining whether online text was produced by artificial intelligence previously required a complex autopsy. Under the new European framework, each generated passage will effectively carry its own birth certificate.
How invisible AI text marking works
The tracking mechanism remains entirely invisible to human readers while precisely recording how an artificial intelligence tool selects words to construct a sentence.
The tool leaves subtle clues within vocabulary choices, sentence structure and punctuation patterns, providing a set of key style signatures that other AI detection tools can read automatically.
ChatGPT, developed by San Francisco based artificial intelligence research firm OpenAI, alongside Anthropic's Claude and Google's Gemini, rely on large language models to generate human-like writing. These automated systems process user prompts to draft articles, essays, customer emails and creative text within seconds.
Text watermarking relies on mathematical algorithms that subtly bias how vocabulary is selected during automated text generation. By weighting specific word choices, the software embeds a unique statistical fingerprint into the text while keeping the final output completely natural for human readers.
Major generative platforms including ChatGPT, Claude and Gemini have already begun deploying the underlying marking technology across their systems ahead of the December compliance deadline.
However, Mbida highlighted a major drawback of the new system, explaining that any text created prior to the deployment of this marking tool will remain impossible for the software to detect.
Access limits and primary uses
The overall reliability of the new detection system remains difficult to evaluate because the detection tools are not currently accessible to the general public.
At present, access to the detection tools is restricted to technology experts, academic research organizations and selected news media outlets testing the capability.
The system also faces technical constraints linked to text length. Detection rates drop significantly when analyzing shorter passages, meaning very brief text snippets may escape identification altogether.
TF1, France's oldest and largest commercial television network, broadcasts Bonjour La Matinale TF1 as its daily morning news and current affairs show. Technology contributor Anicet Mbida regularly presents segments explaining digital policies and emerging technology trends.
The European Union has consistently led international efforts to regulate emerging digital technologies. European legislative frameworks mandate strict transparency standards for synthetic media across all 27 member states, seeking to help consumers distinguish between human-written and machine-generated content.
The rapid expansion of consumer AI tools has heightened concerns over automated content farms, fake consumer reviews on e-commerce platforms and synthetic misinformation campaigns designed to deceive online audiences.
The primary objective of the European Union content transparency standard is to restore trust in digital media and help platforms spot widespread online spam.
Regulators expect the new watermarking technology to identify automated junk emails, fabricated news articles and fake customer reviews published across the internet.
