Key Takeaways
- OpenAI has introduced an invisible, machine-readable watermark in text output for ChatGPT and Codex.
- The watermark is being rolled out initially to users in the European Union to comply with AI Act requirements.
- OpenAI claims textGrain watermarking matches or exceeds other approaches like Google DeepMind's SynthID.
OpenAI has announced the implementation of an invisible, machine-readable watermark in text output for its AI platforms, ChatGPT and Codex, for users in the European Union. This move is part of efforts to comply with the European Union's AI Act, which mandates certain measures to ensure transparency and accountability in AI systems.
According to OpenAI, the new watermarking system, named textGrain, is designed to be undetectable to human users while providing a means for machines to identify the source of the text. The company states that textGrain has been tested and found to match or exceed the performance of other watermarking methods, such as Google DeepMind's SynthID, which is also used by Anthropic in its AI systems.
OpenAI has released scores from AI benchmarks that demonstrate the performance of watermarked and unwatermarked text. These results indicate that the watermarking does not significantly impact the quality or functionality of the text output, ensuring that users will not experience any noticeable changes in their interactions with ChatGPT and Codex.
While the introduction of this watermarking system is a significant step for OpenAI, it has not been universally welcomed. Some users and privacy advocates have expressed concerns about the potential for increased surveillance and loss of anonymity in online communications. However, OpenAI maintains that the watermarking is necessary to meet regulatory requirements and ensure the integrity of its AI systems.
The rollout of textGrain watermarking to EU users is the first step in a broader strategy by OpenAI to comply with the AI Act. The company is expected to expand the implementation of this technology to other regions and platforms in the coming months, as it continues to adapt to the evolving landscape of AI regulation.
OpenAI has stated that the textGrain watermarking does not guarantee complete anonymity or prevent the detection of AI-generated content. However, the company believes that the system provides a balanced approach to maintaining the utility of its AI platforms while adhering to legal and ethical standards.
