EU AI Act enters force, mandating watermarks for AI content

EU AI Act AI watermarking content provenance generative AI regulation digital transparency

Key provisions of the EU AI Act officially came into force on August 2, 2026, establishing a regulatory framework for the safe and transparent use of artificial intelligence within the European Union. The law mandates that users must be notified when interacting with AI chatbots and requires machine-readable watermarks for AI-generated or significantly modified visual, audio, and text content to ensure transparency.

Major technology providers are implementing various technical solutions to meet these requirements. Anthropic has announced the deployment of text watermarking for its Claude models, using specific word selection patterns that are detectable by authorized parties. Google has integrated its SynthID framework into Gemini, while OpenAI is utilizing a combination of C2PA metadata and SynthID pixel watermarking for images. Meta is also applying both C2PA metadata and deep learning-based watermarking to its AI-generated content.

These measures aim to mitigate risks such as deepfakes and misinformation. However, the implementation has sparked debate, including the rapid development of open-source tools designed to remove these digital watermarks and criticisms regarding the potential impact of watermarking on the quality and accuracy of AI-generated text.

Entities

Anthropic · European Union · Google · Meta · OpenAI

Claims

What the coverage asserts, and how many sources carry each claim.

[● 2 SOURCES] AI-generated or significantly modified visual and audio content must now include machine-readable watermarks. dulist.hr · www.infoq.cn
[● 2 SOURCES] The AI Act mandates transparency, requiring systems to notify users when they are interacting with AI chatbots. dulist.hr · bairesparatodos.com.ar
[○ 1 SOURCE] In 2025, 20% of EU companies used artificial intelligence, up from 13% the previous year. metro-portal.hr
[● 2 SOURCES] Key provisions of the EU AI Act officially came into force on August 2, 2026. dulist.hr · www.infoq.cn
[○ 1 SOURCE] The UNDP has opened applications for its 2026 Digital, AI, and Innovation Internship program until September 30, 2026. www.mreza-mira.net
[○ 1 SOURCE] 53% of companies in the EU used cloud services. metro-portal.hr
[● 2 SOURCES] The EU has banned AI systems that categorize individuals based on biometric data to determine race, political, or religious beliefs. dulist.hr · bairesparatodos.com.ar
[○ 1 SOURCE] Meta applies both C2PA metadata and deep learning image watermarking to its AI content. www.infoq.cn
[○ 1 SOURCE] Google has integrated its SynthID framework into Gemini production infrastructure to provide content provenance. www.infoq.cn
[● 2 SOURCES] Anthropic is implementing text watermarking in its Claude models to comply with EU transparency regulations. www.infoq.cn · japan.zdnet.com
[○ 1 SOURCE] OpenAI uses a combination of C2PA metadata and SynthID pixel watermarking for image provenance. www.infoq.cn