As OpenAI continues to push the boundaries of artificial intelligence, the company has announced that it will be adding invisible watermarks to text generated by its popular ChatGPT and Codex models in the European Union. This move aims to combat the growing concern of AI-generated content being used for malicious purposes, such as spreading misinformation or propaganda.
The watermarks will not be visible when users read or copy the text, but they will create a statistical pattern that can later be detected using OpenAI’s new “textGrain” technology. While this may sound like a straightforward solution to identifying AI-generated content, it’s worth noting that the watermark is far from foolproof. In fact, OpenAI’s own tests have shown that even minor editing of the text can significantly reduce its ability to detect the watermark.
For example, in one test, replacing just 10% of the words with synonyms reduced detection rates from around 92% to 66%. Furthermore, the watermark may not be effective for certain types of content, such as mathematics, where the model has limited flexibility in choosing different words. OpenAI has warned that a detected watermark does not necessarily prove human authorship, and that text generated by its tools may still be too short or edited to be reliably detected.
The addition of watermarks is part of a broader effort by OpenAI to promote transparency and accountability in the use of AI-generated content. While it remains to be seen how effective this technology will be in practice, it’s an important step towards addressing the growing concern of AI-powered misinformation campaigns.
OpenAI has also announced that it will not make watermarking a global default, at least not yet. Instead, API developers worldwide can opt-in to watermarking for supported models, but it remains disabled by default. This means that users who want to take advantage of this feature will need to explicitly enable it through their API settings.
In related news, OpenAI has also announced the availability of its watermark detector tool, although access is initially limited to approved researchers and expert organizations. This tool promises to help identify AI-generated content with a high degree of accuracy, but its effectiveness will depend on various factors, including the quality of the text and the level of editing that has been applied.
As AI-powered attacks continue to evolve and become more sophisticated, it’s essential for users to be aware of these developments and take steps to protect themselves. While OpenAI’s watermarking technology is an important step towards promoting transparency in AI-generated content, it’s just one part of a broader effort to address the challenges posed by AI-powered threats.
To stay ahead of the curve, it’s crucial to understand how AI-powered attacks work and what you can do to mitigate their impact. Consider taking advantage of tools like watermarking technology to identify AI-generated content, but also be aware that this is just one aspect of a larger strategy for protecting yourself from AI-powered threats.
Source: Bleeping Computer — 2026-10-05