Anthropic Introduces AI Watermarking for Claude and Updates Auto Mode Settings

AI giant Anthropic has started watermarking outputs from its Claude chatbot. The company is now embedding invisible code into the text to mark that it is AI-generated. According to a report by TechCrunch, this move from Anthropic comes in response to the EU AI Act’s Transparency Code, which requires companies to label AI-generated or edited content in ways detectable by computer systems. The regulators see this as a step toward accountability in AI use across Europe.

While European regulators welcomed the change, some users have voiced frustration. On Reddit, critics argued the watermarking system unfairly exposes everyday use cases — from students reorganizing paragraphs to journalists summarizing transcripts. One poster described it as a “digital tattoo on the forehead,” while another called the practice “unethical” and “disgusting,” claiming Claude is merely a tool that assists human effort. Not all users agreed with the outrage. Many defended the watermarking as a sensible safeguard, pointing out that it helps detect AI-generated material in contexts where risks are high. “There is literally no good argument for why this isn’t a good idea,” one commenter wrote, adding that the only reason to oppose it would be to mislead others.

Meanwhile, in related news recently Anthropic announced that its auto mode will become the default setting for its Claude Code platform starting August 14. This change will affect the Pro, Max and Team accounts. With this new change in place, the developers will no longer need to manually approve each actin that AI takes during programming tasks. Instead, Claude Code will automatically process unless an action is deemed ‘irreversible, destructive, or aimed outside your environment’. The company first introduced auto mode in March as a test feature designed to balance speed and control. In its latest announcement, Anthropic said testing showed auto mode to be safer than manual review — catching 89% of harmful actions compared to just 13.6% detected by human reviewers. The company noted that users tend to approve nearly all permission prompts manually, making automated oversight more effective.