Unverified50% confidenceFactExact time
The watermarking solution in the biological domain directly reuses the open-source SynthID text library code, so model developers can use it without coordinating with DeepMind.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
UnverifiedThe author conducted a continuation experiment using Huawei Cloud CodeArts to verify that LLMs generate content rather than retrieve answers.66% similarUnverifiedSynthID通过在扩散模型采样过程中干预来嵌入水印，对于文本则通过调整token概率分布嵌入水印64% similarUnverifiedSynthID embeds imperceptible statistical signals into the bit or semantic layer of content at the generation stage rather than relying on appended metadata, enabling the watermark to remain detectable after compression, editing, or format conversion.64% similarUnverifiedWatermarking technology requires active cooperation from the model provider, and the watermark signal can be effectively removed through paraphrase attacks (rewriting the text with another model).63% similarUnverifiedScenario-differentiated watermarking may become a trend, applying stronger watermarks to purely generative content while adopting lighter or selective strategies for critical tokens involving tool calls and code execution.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/972907API
curl https://kongchang.com/api/v1/knowledge/claims/972907MCP
get_claim(id=972907)