OpenAI is rolling out an invisible, machine-readable watermark for text produced by ChatGPT and Codex, beginning with users in the European Union. According to reporting by The Verge, OpenAI said the change is intended to help meet requirements linked to the EU AI Act.

For businesses using generative AI in marketing, support, coding or internal documentation, the announcement is notable because it points to a more formal compliance layer around AI-generated content in Europe. It also highlights a practical limitation: watermarking may help identify AI-generated text, but it is not a guarantee of detection in every case.

What happened

OpenAI said it is adding a text watermark to output from ChatGPT and Codex for EU users first. The watermark is described as invisible to readers but readable by machines. The company reportedly said its approach, called textGrain, performed similarly to or better than other text watermarking methods in its own testing.

The development follows similar moves by other AI providers. The Verge reports that Anthropic previously announced text watermarking based on technology from Google DeepMind, also in the context of EU AI Act requirements.

Based on the source, OpenAI also shared benchmark results indicating similar performance between watermarked and non-watermarked text. However, OpenAI reportedly noted that the system does not guarantee reliable detection in all circumstances.

Why it matters for European businesses

This matters because many European companies are moving from experimental AI use to operational use cases such as content production, customer service, software development and workflow automation. As AI adoption becomes more formal, providers are increasingly building in compliance-oriented features for the European market.

Text watermarking could affect businesses in several ways:

• Content governance: Companies may need clearer internal rules for when AI-generated text can be published, edited or reused.
• Client and customer transparency: Some organisations may choose to document when content was AI-assisted, especially in regulated or trust-sensitive sectors.
• Procurement and vendor review: Businesses evaluating AI tools may increasingly ask whether outputs include traceability, watermarking or other provenance features.
• Risk management: Watermarking may support internal monitoring, but it should not be treated as a complete control against misuse, plagiarism concerns or policy violations.

For European businesses, the broader signal is as important as the feature itself: AI systems offered in the EU may diverge from versions available in other markets as providers adapt products to European regulatory expectations.

Who may be affected

• Marketing teams and agencies using ChatGPT for website copy, campaign drafts, product descriptions, email content or SEO support.
• Software and IT teams using Codex or similar tools to generate code, documentation or technical text.
• E-commerce businesses creating large volumes of catalogue, support and transactional content with AI assistance.
• Founders and SMEs adopting AI tools without a formal governance framework yet.
• Regulated businesses that need stronger auditability, documentation or controls around automated content generation.

What companies should consider

• Review AI usage policies: Make sure staff know which business content can be AI-generated, which requires human review and which should be excluded from automated drafting.
• Do not rely on watermarking alone: If your business needs provenance or compliance evidence, treat watermarking as one signal rather than a complete detection or enforcement mechanism.
• Ask vendors about EU-specific features: AI providers may apply different controls in Europe. Procurement, legal and IT teams should understand what changes apply to EU users.
• Update client-facing workflows where needed: Agencies, publishers and service providers may want to decide whether and how to disclose AI-assisted content creation.
• Assess higher-risk use cases separately: Content used in regulated communications, legal documentation, financial information or sensitive customer interactions may need stricter review and approval processes.

The immediate impact is not that every AI-generated text will become easily identifiable, but that major providers are now embedding compliance-related controls directly into products used by European businesses. For companies scaling AI use, governance and documentation are becoming part of everyday operations, not just a legal side issue.