Coders Say They Already Found Workarounds to Claude’s Invisible Watermarks
Coders have found workarounds to remove invisible watermarks from AI-generated content, raising concerns about the effectiveness of the EU's AI Act. Developer Guillaume Meyer's code to remove watermarks has gone viral on GitHub, with many incorporating the technology into…
Intelligence analysis by Llama
Coders have found ways to remove invisible watermarks from AI-generated content, sparking concerns about the EU's AI Act. Developer Guillaume Meyer's code has gone viral on GitHub, with many incorporating the technology into their own projects.
Imagine you're writing a story, but someone else is secretly adding a special code to it that only a computer can see. This is what's happening with AI-generated content, where invisible watermarks are being added to make it detectable as AI-generated. But some clever coders have found ways to remove these watermarks, raising questions about the effectiveness of the EU's AI Act.
Analysis
EU AI Act and Invisible Watermarks
The European Union's AI Act, which came into effect earlier this month, requires model providers like Anthropic and OpenAI to label synthetic audio, image, video, or text so that this material can be detected by a machine as AI-generated. The rules stipulate that providers cannot market circumvention tools, but there is no legal restriction on independent tools. This has led to a cat-and-mouse game between model providers and developers who are trying to find ways to remove the invisible watermarks.
Watermarking and Quality of AI Responses
Anthropic watermarks text invisibly by leaving a pattern in Claude's choice of words and phrases that is indiscernible to a human reader but would be detectable by a machine that knows how to look for it. Some users are concerned that this will degrade the quality of Claude's responses, though Anthropic insists this won't be the case. The technique, called SynthID, was developed by Google, which has been using it to watermark its AI-generated content since 2023.
Workarounds and Removal Tools
Coders have developed various workarounds and removal tools to counter the invisible watermarks. Developer Guillaume Meyer's code to remove watermarks has gone viral on GitHub, with many incorporating the technology into their own projects. Software engineer Erik Hughes took 15 minutes to knock up a tool with Claude that removes invisible and look-alike characters, reorders sentences within paragraphs, and swaps several words for synonyms. Leon Chlon, a Visiting Fellow at the University of Oxford, says the watermarks can be removed by condensing Claude's response, translating it into a dialect like Arabic, which has very different semantics compared to English, and then translating it back.
Implications and Future Developments
The discovery of workarounds to remove invisible watermarks from AI-generated content raises concerns about the effectiveness of the EU's AI Act and the potential for misuse of AI technology. Anthropic itself acknowledged that heavily edited, paraphrased, or translated content might not carry a watermark. In a statement to WIRED, a spokesperson for Anthropic said: "We're adding marking to Claude's output to comply with the EU AI Act, and other labs are taking similar steps. It's hard to identify AI-generated text, and this gives people better tools for identification." Anthropic says it's working out how to implement watermark detection for text and plans to release a tool to do so soon—at which point developers will finally be able to see whether their methods are foolproof.
Key points
- Coders have found workarounds to remove invisible watermarks from AI-generated content.
- Developer Guillaume Meyer's code to remove watermarks has gone viral on GitHub.
- Anthropic watermarks text invisibly by leaving a pattern in Claude's choice of words and phrases.
- Some users are concerned that this will degrade the quality of Claude's responses.
- Coders have developed various workarounds and removal tools to counter the invisible watermarks.
The discovery of workarounds to remove invisible watermarks from AI-generated content could lead to a more nuanced understanding of the EU's AI Act and the potential for misuse of AI technology. This could lead to more effective regulations and better tools for identifying AI-generated content.
The discovery of workarounds to remove invisible watermarks from AI-generated content could lead to a cat-and-mouse game between model providers and developers, potentially undermining the effectiveness of the EU's AI Act and the potential for misuse of AI technology.
