AI Act: shortly after mandatory labeling of AI content is introduced, tools emerge to circumvent it
Shortly after mandatory labeling of AI content under Article 50 AI Act was introduced, tools emerged to remove it. According to the source, Grok from Elon Musk ignores labeling, which may paradoxically favor models beyond the reach of EU regulation, such as Chinese models.
Shortly after mandatory labeling of AI content under Article 50 AI Act came into force (2 August 2026), tools emerged that can remove traces of artificial intelligence from text, according to E15.cz. The website describes how text labeling is intended to work through a combination of inserting invisible characters and selecting words non-randomly, while new “removers” can erase invisible characters or regenerate text that uses non-random word selection. Anthropic stated in response that the Claude model does not insert any invisible characters and only selects words non-randomly, without this being intended to change the meaning of the text.
According to E15.cz, the Gemini model from Google has been labeling content for some time, the Claude model from Anthropic began labeling at the beginning of August, while Grok from Elon Musk apparently ignores the obligation. The website highlights a possible unintended consequence of the regulation: because Chinese models do not comply with European rules, users may switch to them out of concern about incorrect labeling or a loss of quality, which would paradoxically favor models beyond the reach of AI Act and reduce the productivity of users who comply with European rules.
This is also related to the so-called boomerang effect described by The Conversation: if people become accustomed to AI content carrying a label, they may mistakenly assume that unlabeled content is automatically authentic—even though actors who deliberately seek to deceive through content do not label it. According to the website, relying on labels alone may therefore be insufficient, and content needs to be verified in other ways as well.
As a reminder, Article 50 AI Act generally requires companies to clearly inform users when they are communicating with chatbots or AI agents, requires deepfakes and AI-generated content to carry machine-readable labels, and requires advertising visuals or product photographs created by AI to be appropriately labeled as well. Violations can result in fines of up to 15 million euros or 3 % of the company's worldwide turnover.
Why it matters
For companies, this poses a competitive risk: those that diligently comply with the obligations under Article 50 may be disadvantaged compared with competitors using models that ignore or evade the rules (according to the source, Grok or Chinese models), especially as tools emerge that allow mandatory markers to be removed from text. For individuals, this means that the presence or absence of official AI labeling alone may not be a reliable signal of the authenticity of content.
What was added since the original report
Verified updates
-
Tools exist that can remove or cover up mandatory AI labels; Labeling of AI content works through invisible characters and non-random word selection; AI Act regulation may become ineffective because of circumvention techniques; Grok from Elon Musk ignores mandatory labeling; The regulation may paradoxically favor Chinese models outside the scope of EU rules
- Tools exist that can remove or cover up mandatory AI labels
- Labeling of AI content works through invisible characters and non-random word selection
- AI Act regulation may become ineffective because of circumvention techniques
- Grok from Elon Musk ignores mandatory labeling
- The regulation may paradoxically favor Chinese models outside the scope of EU rules
-
Specific identification of the obligation as Article 50 AI Act; Specification of the technical requirement: mandatory icons for labeling; Specification of the technical requirement: machine-readable watermarks; Specific implementation example: Anthropic will add watermarks to Claude in the EU
- Specific identification of the obligation as Article 50 AI Act
- Specification of the technical requirement: mandatory icons for labeling
- Specification of the technical requirement: machine-readable watermarks
- Specific implementation example: Anthropic will add watermarks to Claude in the EU
-
There is a Code of Practice (Code of Practice) on transparency of AI-generated content that puts regulatory rules into practice; Over 180 organizations have signed the Code of Practice on transparency of AI content; The aim of the rules is to reduce fraud and manipulation and help people make informed decisions; Machine-readable markers make automatic detection of AI-generated content easier; The new rules provide businesses with clear legal obligations and practical means of demonstrating compliance
- There is a Code of Practice (Code of Practice) on transparency of AI-generated content that puts regulatory rules into practice
- Over 180 organizations have signed the Code of Practice on transparency of AI content
- The aim of the rules is to reduce fraud and manipulation and help people make informed decisions
- Machine-readable markers make automatic detection of AI-generated content easier
- The new rules provide businesses with clear legal obligations and practical means of demonstrating compliance
-
More than a third of companies address AI only superficially; The regulation was amended through the 'digital omnibus'; The amendment was published at the end of July
- More than a third of companies address AI only superficially
- The regulation was amended through the 'digital omnibus'
- The amendment was published at the end of July
Two audiences, two different impacts
What this means
For individuals
The presence or absence of mandatory labeling of AI content is no longer a reliable guide, because according to sources, tools exist that can remove such markers, and some models (Grok) ignore them according to information from the source.
For a business
Companies that comply with the obligations under Article 50 AI Act face the risk that their properly labeled AI content will be disadvantaged compared with competitors using models beyond the reach of the regulation (according to the source, Grok or Chinese models), because tools have also emerged that can remove mandatory markers from text.
Risks and complianceCheck the original
Event sources
confirmed by 3 independent sources · 4 publishers, 3 independent. We count feeds from the same owner only once.