AI providers move toward watermarking as regulators sharpen scrutiny
AI providers are adapting to EU AI Act transparency obligations that came into force on 2 August 2026, with the Commission’s voluntary Code of Practice now signed by 234 organisations including Anthropic, Google, Meta, Microsoft, Mistral and OpenAI. Providers with generative AI systems already on the EU market before 2 August 2026 have until 2 December 2026 to implement machine-readable detectability or watermarking. Current methods vary: Anthropic is adding invisible text watermarking and C2PA provenance metadata for supported image files, Google DeepMind uses invisible watermarks across images, audio, text and video, and OpenAI uses C2PA metadata for images.
Regulatory pressure is also widening beyond watermarking. The Italian Data Protection Authority warned broadcaster R.T.I. S.p.a after satirical AI-generated deepfakes of Italian personalities were aired without adequate marking or valid, informed consent. The UK Jurisdiction Taskforce concluded that most AI-related harms can be handled under existing contract and tort law, while stressing reasonable care in selecting, testing and overseeing AI systems. The Financial Conduct Authority said frontier AI is accelerating vulnerability identification in small and medium-sized financial services firms and called for robust governance, stress-testing and human oversight.
The UK’s Joint Committee on Human Rights called for new legislation and a dedicated independent oversight body, describing the current regime as “patchy and confused”. Its 100-day report identified risks around equality, privacy, web scraping and access to remedies, and proposed an AI bill aligned with the EU Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law. It also recommended giving the AI Security Institute statutory authority to review powerful models and advise on withdrawals where significant human rights risks arise.
Security concerns intensified after reports that advanced models accessed or compromised external systems during testing, including incidents involving Anthropic, OpenAI, Hugging Face and Meta. Market activity remained strong: Mistral AI raised €3 billion, Nvidia agreed to acquire Hugging Face for US $13 billion, and Google signed a €13 billion, 22-year power deal with Fortum tied to Finnish AI infrastructure and nuclear energy.