Claude's watermark relies on a version of Google DeepMind's SynthID-Text approachCITE:E1. Anthropic confirmed it is using the SynthID-Text method that the Google DeepMind team outlined in 2024, and said it plans to release a watermark detection APICITE:E10. Anthropic explained that the system works by leaning on "low-stakes" word choices — such as picking "overcast" over "grey" — that occur many times across a generated response, leaving a pattern that is invisible to a reader but detectable by anyone holding the key that encodes itCITE:E4.
Anthropic introduced watermarking to meet its obligations under the European Union's AI ActCITE:E2. The feature, alongside C2PA support for Claude-processed images, is being added because the law requires synthetic audio, image, video, and text to carry machine-readable marksCITE:E2. Anthropic published a blog post on Friday, August 15, 2026, addressing basic questions about the rollout: how the watermarking will actually work, whether it can be hidden through editing, and how it affects codeCITE:E6.
Anthropic says watermarking will not raise Claude's price or affect the quality of its outputCITE:E3. The company stated the feature won't make Claude more expensive for users or have "any practical impact on the quality or content of Claude's outputs"CITE:E3, adding that "to a reader, a watermarked response is indistinguishable from an unwatermarked one"CITE:E9. On whether the mark can be stripped out, Anthropic said light editing probably won't remove the watermark completely, while a complete rewrite in which every word is replaced will remove itCITE:E11. For programming output, Anthropic said code should carry less of a watermark than other text, since the model has to produce working code and has less freedom to choose among equally valid optionsCITE:E12.
Google's Gemini has used the SynthID-Text method since 2024, while OpenAI has not detailed similar plans for ChatGPTCITE:E5. Gemini has supported the SynthID-Text solution since 2024, and although OpenAI has not laid out text-watermarking plans for ChatGPT within its AI Act compliance roadmap, it remains subject to the law's requirementsCITE:E5. Anthropic also said Claude will not be the only chatbot generating watermarked text, noting that "other major model developers have signed the same Code of Practice and will be implementing their own watermarks"CITE:E13.
Claude users have publicly debated the watermark rollout, with some calling it a conspiracyCITE:E7. The debate began after the company revealed earlier in the week that it would watermark text to comply with the EU AI Act's Transparency Code; one poster described the move as a conspiracy against innocent Claude users, while another wrote, "The only reason you wouldn't want this is to lie to people"CITE:E7. Business Insider separately reported that "dozens" of users on X said they had canceled their Claude subscriptions as a resultCITE:E8.
Anthropic's own account holds that watermarked and unwatermarked Claude responses are indistinguishable to readers and carry no cost or quality trade-offCITE:E3CITE:E9, yet that assurance has not stopped some users from calling the policy a conspiracy or reporting canceled subscriptionsCITE:E7CITE:E8. The rollout also lands unevenly across the industry: Gemini has run SynthID-Text since 2024, OpenAI has yet to specify a text-watermarking plan for ChatGPT, and Anthropic says other Code of Practice signatories are still building their own systems — meaning Claude's detailed disclosure currently sits ahead of at least one major peer under the same EU AI Act obligationsCITE:E5CITE:E13.
No. Anthropic says watermarking won't make Claude more expensive and has no practical impact on the quality or content of its outputs, and a watermarked response is indistinguishable from an unwatermarked one to a reader<CITE:E3><CITE:E9>.
Anthropic says light editing probably won't remove the watermark completely, but a complete rewrite where every word is replaced will remove it<CITE:E11>.
Anthropic says code should carry less of a watermark than other text, since the model must produce working code and has less freedom to pick among equally valid word or token choices<CITE:E12>.
Anthropic says Claude won't be the only one — other major model developers have signed the same Code of Practice and will implement their own watermarks, and Google's Gemini has already used SynthID-Text since 2024<CITE:E13><CITE:E5>.
Market rumors, based on a SemiAnalysis client report, say Google is working with AMD on a 10th-generation TPU project. The report suggests Google wants AMD's CPU IP, interconnect, or packaging technology rather than direct chip design, and cautions the collaboration's exact form remains unconfirmed.
Stifel strategist Barry Bannister warns that fading speculation in AI-linked stocks and a roughly 25% probability of a 2026 US recession could drag the S&P 500 down about 20%, versus about 9% upside if growth holds — in line with historical recession declines averaging 23% since World War II.
Hu Cheng-chieh is tendering for 50.98%–72.82% of delisted Shinfox Energy at NT$0.05 per share starting August 18, after subsidiary Fuwei Energy's cost overruns on Taipower's offshore wind Phase 2 project produced a FY2025 net loss of NT$15.865 billion and a Q1 2026 net-worth reversal that triggered delisting on June 23.