#  AI Watermark Detection Is Theater 

 

  ![AI Watermark Detection Is Theater](https://cdn.richeyweb.com/images/articles/ai-watermark-detection-is-theater/ai-watermark-detection.webp)    I spent a little time looking into whether a browser plugin could let someone highlight any block of text and check it for an AI watermark. The idea seemed straightforward enough: if the labs are embedding detectable signals in their outputs, a user should be able to surface them. It didn’t take long to hit the wall.

## The public story and why it fails

Detection for these keyed statistical watermarks requires the secret keys, the exact tokenizer, and the generation configuration used at sampling time. Those stay with the platform that applied the mark. OpenAI can only detect OpenAI-watermarked text (to the extent it watermarks at all). Google can only detect Gemini’s. Anthropic can only detect Claude’s. There is no shared detector and no practical way for a third party to run one. Google’s own [Responsible GenAI Toolkit documentation](https://ai.google.dev/responsible/docs/safeguards/synthid "Google SynthID") makes the dependency plain. That ended the plugin idea. It also made the larger point hard to miss: AI watermark-based detection is useless for anyone except the providers themselves.

The public story around these systems is that they give the rest of us a way to identify machine-generated text. Labs have deployed or announced versions under regulatory pressure, including EU AI Act transparency rules. Google’s approach is described in the [Nature paper on SynthID-Text](https://www.nature.com/articles/s41586-024-08025-4 "Scalable watermarking for identifying large language model outputs") and the [open-source notes](https://huggingface.co/blog/synthid-text "Introducing SynthID Text"). Anthropic has outlined a [similar method for Claude](https://www.anthropic.com/news/claude-text-watermark "How Claude’s text watermark works"). The mechanism is a subtle, key-dependent bias introduced during sampling. A detector holding the matching key can later measure it. Readers are not supposed to notice.

### What does our search overlord say about AI watermarks?

Google’s ranking guidance has never treated pure AI detection as the signal that matters. Its position since early 2023, restated in the October 2026 updates, is that appropriate use of AI is not against the guidelines. What it targets under the scaled content abuse policy is mass production of unoriginal pages whose main purpose is manipulating rankings rather than helping users - no matter how the pages were made. The same updates said it is critical to manually fact-check AI output before publishing because models can hallucinate. Quality rater guidelines tell raters to score low-effort, low-originality AI content as Lowest quality. There is no public evidence that commercial detector scores are used as ranking factors. See the [generative AI content guidance](https://developers.google.com/search/docs/fundamentals/using-gen-ai-content "Google Search's guidance on using generative AI content on your website"), coverage of the [October 2026 changes](https://www.searchenginejournal.com/google-fact-check-ai-content-before-publishing/591782/ "Google Tells Sites To Fact-Check AI Content Before Publishing"), and explanations of [scaled content abuse](https://sumukhai.io/insights/does-google-penalize-ai-content "Does Google Penalize AI Content? What Google's Guidance Actually Says").

Even for the platform that holds the keys, the signal is easy to break. It is a statistical bias in token probabilities, not a durable signature. Light rewording through another model, commercial humanizers, multi-pass paraphrasing, translation, or ordinary editing all disrupt it. The humanizer market exists because the patterns are fragile. Benchmarks such as [HumanizerBench](https://humanizerbench.com/ "The benchmark for AI humanizers") track the back-and-forth, and write-ups on detector limits (for example [this one](https://www.authmarks.com/learn/how-ai-detectors-work-and-why-they-break "How AI text detectors actually work — and the three ways they break")) show how quickly scores collapse under rewriting. Open-sourcing the algorithm does not help outsiders when the production keys stay private.

Nobody builds serious public systems around this because the secrets are not shared and the mark does not survive normal editing. That leaves the providers as the only parties who can actually use the mechanism - and only for purposes that do not require anyone else to detect it.

## The AI watermark internal customer: radioactivity

The research literature has looked at one such purpose for years not under the name AI watermark, but under the name watermark radioactivity. When a model is trained or distilled on a substantial volume of watermarked text, residual bias from the original sampling can remain detectable in the resulting model’s outputs, or in some accounts as a key-specific geometric signal in parameter space that sits apart from ordinary task fine-tuning.

Meta’s 2024 paper showed that fine-tuning on watermarked synthetic instructions could be detected at high confidence even when only about 5% of the training data carried the mark ([“Watermarking Makes Language Models Radioactive”](https://arxiv.org/abs/2402.14904 "Watermarking Makes Language Models Radioactive")). Later work treats this as a way to spot unauthorized knowledge distillation: if a competitor scrapes large volumes of API outputs and trains on them, the original lab can later look for the residual. Papers such as [“Can LLM Watermarks Robustly Prevent Unauthorized Knowledge Distillation?”](https://arxiv.org/abs/2502.11598 "Can LLM Watermarks Robustly Prevent Unauthorized Knowledge Distillation?"), work on [Watermark Basins](https://arxiv.org/html/2610.37266v1 "Watermark Basins: Black-Box Detection of Distilled LLMs"), and schemes like [TextSeal](https://arxiv.org/html/2605.12456v2 "TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection") explore the same idea. The original keys are still required, which is fine when the only party that needs to run the test is the one that applied the AI watermark.

This does not require third-party access, a cross-platform scan, or the mark surviving public rewriting of individual documents. A platform can AI watermark its own outputs and later check whether a competitor’s model shows the corresponding residual. It functions as a private tracer for competitive monitoring. The public detection story does not hold; the private one can.

## Limitations that remain

The internal application is still statistical and imperfect. Targeted paraphrasing of the training data, filtering, dilution with non-watermarked text, or certain inference-time neutralization methods can scrub the inherited signal while still transferring capability. Power depends on volume, proportion, and the fine-tuning process. It is evidence, not proof. Attack papers exist (see [unified spoofing and scrubbing work)](https://arxiv.org/abs/2504.17480 "Unified attacks to large language model watermarks: spoofing and scrubbing in unauthorized knowledge distillation"), and more robust variants have been proposed in response.

## What this means

For everyone else, these AI watermarks were never going to solve detection for ranking, moderation, or labeling. Private keys, separate platforms, and easy disruption make public use non-viable. That is why third parties ignore it and why humanizers are common. For the labs, the radioactivity property supplies a monitoring tool that does not require anyone else to be able to see the mark. It is not a hard barrier against distillation, but it is a detectable trace when outputs are used at scale without laundering. That capability is available only to the party that controls the keys.

The public story and the technical constraints do not match. The mechanism that fails for everyone else can still be useful to the platform that applied it. Public detection is theater. The private tracer is the part that was never meant for the audience.



- [      email ](mailto:?subject=AI+Watermark+Detection+Is+Theater&body=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fai-watermark-detection-is-theater)
- [      facebook ](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fai-watermark-detection-is-theater)
- [      x-twitter ](https://twitter.com/intent/tweet?text=AI+Watermark+Detection+Is+Theater%3A+https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fai-watermark-detection-is-theater)
- [      linkedin ](http://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fai-watermark-detection-is-theater&title=AI+Watermark+Detection+Is+Theater&summary=I+spent+a+little+time+looking+into+whether+a+brows...)
- [      pinterest ](http://pinterest.com/pin/create/button/?url=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fai-watermark-detection-is-theater&media=https%3A%2F%2Fcdn.joomla.org%2Fimages%2Fjoomla-org-og.jpg&description=AI+Watermark+Detection+Is+Theater)
 


 

   [  Next article: The AI Prompt That Knew Too Much  The AI Prompt That Knew Too Much  ](/blog/development/the-ai-prompt-that-knew-too-much)  

##### We Value Your Privacy

 

We use cookies to enhance your experience and for traffic analysis. By continuing to visit this site you agree to our use of cookies.

[Privacy Policy](/privacy-policy)

 Details 

###### Google Tag Manager Items

- Ad Storage
- Ad User Data
- Ad Personalization
- Analytics Storage
- Functionality Storage
- Personalization Storage
- Security Storage
 
 

 

 

 

 

 Decline Accept
```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://www.richeyweb.com/#organization","name":"RicheyWeb","url":"https://www.richeyweb.com/","logo":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/logo/richeyweb112.svg","contentUrl":"https://www.richeyweb.com/images/logo/richeyweb112.svg","width":{"@type":"QuantitativeValue","value":112,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":112,"unitCode":"PX"},"@id":"https://www.richeyweb.com/#logo"},"image":{"@id":"https://www.richeyweb.com/#logo"},"sameAs":["https://x.com/ComRicheyweb","https://www.facebook.com/RicheyWebDev/","https://www.youtube.com/channel/UCxnVG8BwOvQRO7hVqNX7T2g","https://community.joomla.org/service-providers-directory/listings/115:richeyweb.html"],"description":"RicheyWeb is a custom software developer specializing in Joomla extensions.","ContactPoint":[{"@type":"ContactPoint","url":"https://www.richeyweb.com/contact-us","telephone":"903-873-8460","contactType":"Owner/Administrator","areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"availableLanguage":"en"},{"@type":"ContactPoint","url":"https://www.richeyweb.com/bugs","telephone":"903-873-8460","contactType":"Technical Support","areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"availableLanguage":"en"}],"knowsAbout":["Computer programming",{"@type":"Thing","name":"Computer programming","sameAs":["https://en.wikipedia.org/wiki/Computer_programming","https://www.wikidata.org/wiki/Q80006","https://g.co/kg/m/01mf_"]},"PHP",{"@type":"Thing","name":"PHP","sameAs":["https://en.wikipedia.org/wiki/PHP","https://www.wikidata.org/wiki/Q59","https://g.co/kg/m/060kv"]},"JavaScript",{"@type":"Thing","name":"JavaScript","sameAs":["https://en.wikipedia.org/wiki/JavaScript","https://www.wikidata.org/wiki/Q2005","https://g.co/kg/m/02p97"]},"arduino","Computer forensics",{"@type":"Thing","name":"Computer forensics","sameAs":["https://en.wikipedia.org/wiki/Computer_forensics","https://www.wikidata.org/wiki/Q878553","https://g.co/kg/m/02wxbd"]},"White hat",{"@type":"Thing","name":"White hat","sameAs":["https://en.wikipedia.org/wiki/White_hat_(computer_security)","https://www.wikidata.org/wiki/Q7995625","https://g.co/kg/m/03ns_5"]},"Search engine optimization",{"@type":"Thing","name":"Search engine optimization","sameAs":["https://en.wikipedia.org/wiki/Search_engine_optimization","https://www.wikidata.org/wiki/Q180711","https://g.co/kg/m/019qb_"]},"Search engine marketing",{"@type":"Thing","name":"Search engine marketing","sameAs":["https://en.wikipedia.org/wiki/Search_engine_marketing","https://www.wikidata.org/wiki/Q846132","https://g.co/kg/m/06mw8r"]},"Digital marketing",{"@type":"Thing","name":"Digital marketing","sameAs":["https://en.wikipedia.org/wiki/Digital_marketing","https://www.wikidata.org/wiki/Q1323528","https://g.co/kg/g/122hcnps"]},"Web hosting service",{"@type":"Thing","name":"Web hosting service","sameAs":["https://en.wikipedia.org/wiki/Web_hosting_service","https://www.wikidata.org/wiki/Q5892272","https://g.co/kg/m/014pz4"]},"Email hosting service",{"@type":"Thing","name":"Email hosting service","sameAs":["https://en.wikipedia.org/wiki/Email_hosting_service","https://www.wikidata.org/wiki/Q5368818","https://g.co/kg/m/09w60m"]},"Internet hosting service",{"@type":"Thing","name":"Internet hosting service","sameAs":["https://en.wikipedia.org/wiki/Internet_hosting_service","https://www.wikidata.org/wiki/Q1210425","https://g.co/kg/m/09w5yw"]},"Virtual hosting",{"@type":"Thing","name":"Virtual hosting","sameAs":["https://en.wikipedia.org/wiki/Virtual_hosting","https://www.wikidata.org/wiki/Q588365","https://g.co/kg/m/024mvh"]},"Web performance",{"@type":"Thing","name":"Web performance","sameAs":["https://en.wikipedia.org/wiki/Web_performance","https://www.wikidata.org/wiki/Q7978612","https://g.co/kg/m/0gfj3f1"]},"Web content management system",{"@type":"Thing","name":"Web content management system","sameAs":["https://en.wikipedia.org/wiki/Web_content_management_system","https://www.wikidata.org/wiki/Q45211","https://g.co/kg/m/0615s2"]},"Content management system",{"@type":"Thing","name":"Content management system","sameAs":["https://en.wikipedia.org/wiki/Content_management_system","https://www.wikidata.org/wiki/Q131093","https://g.co/kg/m/0k23c"]},"General Data Protection Regulation",{"@type":"Thing","name":"General Data Protection Regulation","sameAs":["https://en.wikipedia.org/wiki/General_Data_Protection_Regulation","https://www.wikidata.org/wiki/Q1172506","https://g.co/kg/m/0pk_7xs"]},"SERP",{"@type":"Thing","name":"SERP","sameAs":["https://en.wikipedia.org/wiki/SERP","https://www.wikidata.org/wiki/Q2205811","https://g.co/kg/g/11c5szp7kc"]},"Artificial intelligence",{"@type":"Thing","name":"Artificial intelligence","sameAs":["https://en.wikipedia.org/wiki/Artificial_intelligence","https://www.wikidata.org/wiki/Q11660","https://g.co/kg/m/0mkz"]},"Prompt engineering",{"@type":"Thing","name":"Prompt engineering","sameAs":["https://en.wikipedia.org/wiki/Prompt_engineering","https://www.wikidata.org/wiki/Q108941486","https://g.co/kg/g/11p6kpgt_n"]},"E-learning",{"@type":"Thing","name":"E-learning","sameAs":["https://en.wikipedia.org/wiki/E-learning_(theory)","https://www.wikidata.org/wiki/Q182250","https://g.co/kg/g/122czm1f"]},"Sharable Content Object Reference Model",{"@type":"Thing","name":"Sharable Content Object Reference Model","sameAs":["https://en.wikipedia.org/wiki/Sharable_Content_Object_Reference_Model","https://www.wikidata.org/wiki/Q827811","https://g.co/kg/m/06_40"]},"Experience API",{"@type":"Thing","name":"Experience API","sameAs":["https://en.wikipedia.org/wiki/Experience_API","https://www.wikidata.org/wiki/Q7807728","https://g.co/kg/g/1yw9ktxr8"]},"Joomla",{"@type":"Thing","name":"Joomla","sameAs":["https://en.wikipedia.org/wiki/Joomla","https://www.wikidata.org/wiki/Q13167","https://g.co/kg/m/07qb81"]},"Nginx",{"@type":"Thing","name":"Nginx","sameAs":["https://en.wikipedia.org/wiki/Nginx","https://www.wikidata.org/wiki/Q306144","https://g.co/kg/m/02qft91"]},"MySQL",{"@type":"Thing","name":"MySQL","sameAs":["https://en.wikipedia.org/wiki/MySQL","https://www.wikidata.org/wiki/Q850","https://g.co/kg/m/04y3k"]}],"areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"memberOf":["Mensa International",{"@type":"Organization","name":"Mensa International","sameAs":["https://en.wikipedia.org/wiki/Mensa_International","https://www.wikidata.org/wiki/Q184194","https://g.co/kg/m/0140pf"]},"National Rifle Association",{"@type":"Organization","name":"National Rifle Association","sameAs":["https://en.wikipedia.org/wiki/National_Rifle_Association","https://www.wikidata.org/wiki/Q863259","https://g.co/kg/m/0j6f9"]},"CompTIA",{"@type":"Organization","name":"CompTIA","sameAs":["https://en.wikipedia.org/wiki/CompTIA","https://www.wikidata.org/wiki/Q597534","https://g.co/kg/m/040shq"]},"ISFCE LLC",{"@type":"Organization","name":"ISFCE LLC","sameAs":["https://isfce.com","https://g.co/kg/g/11wxm5r0rg"]}],"hasCredential":[{"@type":"EducationalOccupationalCredential","name":"Joomla 3 Certified Administrator","credentialCategory":"Certification","description":"Administrator Exam is the first available Joomla! certification exam","recognizedBy":{"@type":"Organization","name":"Open Source Matters, Inc.","sameAs":["https://en.wikipedia.org/wiki/Open_Source_Matters,_Inc.","https://g.co/kg/g/11f00wvjhz"]},"url":"https://certification.joomla.org/certified-user-directory/michael-richey","about":["Content management system",{"@type":"Thing","name":"Content management system","sameAs":["https://en.wikipedia.org/wiki/Content_management_system","https://www.wikidata.org/wiki/Q131093","https://g.co/kg/m/0k23c"]},"Web content management system",{"@type":"Thing","name":"Web content management system","sameAs":["https://en.wikipedia.org/wiki/Web_content_management_system","https://www.wikidata.org/wiki/Q45211","https://g.co/kg/m/0615s2"]},"Joomla",{"@type":"Thing","name":"Joomla","sameAs":["https://en.wikipedia.org/wiki/Joomla","https://www.wikidata.org/wiki/Q13167","https://g.co/kg/m/07qb81"]}],"educationalLevel":"expert","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/contact/badge.webp","contentUrl":"https://www.richeyweb.com/images/contact/badge.webp","width":{"@type":"QuantitativeValue","value":300,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":86,"unitCode":"PX"},"caption":"Joomla 3 Certified Administrator"}},{"@type":"EducationalOccupationalCredential","name":"Certified Computer Examiner","credentialCategory":"Certification","description":"Internationally recognized computer forensics certifiecation","recognizedBy":{"@type":"Organization","name":"ISFCE LLC","sameAs":["https://en.wikipedia.org/wiki/ISFCE_LLC","https://g.co/kg/g/11wxm5r0rg"]},"url":"https://isfce.com/","about":["Digital forensics",{"@type":"Thing","name":"Digital forensics","sameAs":["https://en.wikipedia.org/wiki/Digital_forensics","https://www.wikidata.org/wiki/Q3246940","https://g.co/kg/m/0cnxzfx"]},"Computer forensics",{"@type":"Thing","name":"Computer forensics","sameAs":["https://en.wikipedia.org/wiki/Computer_forensics","https://www.wikidata.org/wiki/Q878553","https://g.co/kg/m/02wxbd"]},"Mobile device forensics",{"@type":"Thing","name":"Mobile device forensics","sameAs":["https://en.wikipedia.org/wiki/Mobile_device_forensics","https://www.wikidata.org/wiki/Q6887097","https://g.co/kg/m/06zp3tp"]},"Network forensics",{"@type":"Thing","name":"Network forensics","sameAs":["https://en.wikipedia.org/wiki/Network_forensics","https://www.wikidata.org/wiki/Q7001032","https://g.co/kg/m/05pb280"]},"Database forensics",{"@type":"Thing","name":"Database forensics","sameAs":["https://en.wikipedia.org/wiki/Database_forensics","https://www.wikidata.org/wiki/Q5227405","https://g.co/kg/m/0cgqsy"]}],"educationalLevel":"expert","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/contact/isfce-cce2026.webp","contentUrl":"https://www.richeyweb.com/images/contact/isfce-cce2026.webp","width":{"@type":"QuantitativeValue","value":1000,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":1000,"unitCode":"PX"},"caption":"Certified Computer Examiner"}}],"hasOfferCatalog":{"@type":"OfferCatalog","name":"Web Services","itemListElement":[{"@type":"Offer","itemOffered":{"@type":"Service","name":"Hosting"}},{"@type":"Offer","itemOffered":{"@type":"Service","name":"Development"}},{"@type":"Offer","itemOffered":{"@type":"Service","name":"Search Engine Optimization"}}]}},{"@type":"WebSite","@id":"https://www.richeyweb.com/#website","url":"https://www.richeyweb.com/","name":"RicheyWeb","publisher":{"@id":"https://www.richeyweb.com/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://www.richeyweb.com/search?q={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string","valueMaxLength":256,"valueMinLength":2,"valuePattern":"^[A-Za-z0-9\\s]+$"}},"creator":{"@id":"https://www.richeyweb.com/#organization"},"copyrightHolder":{"@id":"https://www.richeyweb.com/#organization"}},{"@type":"WebPage","@id":"https://www.richeyweb.com/blog/development/ai-watermark-detection-is-theater#webpage","url":"https://www.richeyweb.com/blog/development/ai-watermark-detection-is-theater","name":"AI Watermark Detection Is Theater","description":"AI watermark detection only works for the labs that applied it; third parties cannot verify outputs, making public detection claims misleading.","isPartOf":{"@id":"https://www.richeyweb.com/#website"},"about":{"@id":"https://www.richeyweb.com/#organization"},"inLanguage":"en-GB"},{"@type":"Article","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/ai-watermark-detection-is-theater/ai-watermark-detection.webp","contentUrl":"https://www.richeyweb.com/images/articles/ai-watermark-detection-is-theater/ai-watermark-detection.webp","width":{"@type":"QuantitativeValue","value":1712,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":963,"unitCode":"PX"},"caption":"AI Watermark Detection Is Theater","representativeOfPage":true},"headline":"AI Watermark Detection Is Theater","description":"AI watermark detection only works for the labs that applied it; third parties cannot verify outputs, making public detection claims misleading.","author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"},"datePublished":"2026-11-14T00:00:00+00:00","dateModified":"2026-11-14T00:00:00+00:00","about":["Large language model",{"@type":"Thing","name":"Large language model","sameAs":["https://en.wikipedia.org/wiki/Large_language_model","https://www.wikidata.org/wiki/Q115305900","https://g.co/kg/g/11kc9956b3"]},"Knowledge distillation",{"@type":"Thing","name":"Knowledge distillation","sameAs":["https://en.wikipedia.org/wiki/Knowledge_distillation","https://www.wikidata.org/wiki/Q74253442","https://g.co/kg/g/11j0v0ytdb"]},"Artificial intelligence content detection",{"@type":"Thing","name":"Artificial intelligence content detection","sameAs":["https://en.wikipedia.org/wiki/Artificial_intelligence_content_detection","https://www.wikidata.org/wiki/Q120883091","https://g.co/kg/g/11l2lxvky5"]}],"mentions":["DeepMind",{"@type":"Thing","name":"DeepMind","sameAs":["https://en.wikipedia.org/wiki/Google_DeepMind","https://g.co/kg/m/0_fk4h0"]},"Google",{"@type":"Thing","name":"Google","sameAs":["https://en.wikipedia.org/wiki/Google","https://www.wikidata.org/wiki/Q95","https://g.co/kg/m/045c7b"]},"OpenAI",{"@type":"Thing","name":"OpenAI","sameAs":["https://en.wikipedia.org/wiki/OpenAI","https://www.wikidata.org/wiki/Q21708200","https://g.co/kg/g/11bxc656v6"]},"Anthropic",{"@type":"Thing","name":"Anthropic","sameAs":["https://en.wikipedia.org/wiki/Anthropic","https://www.wikidata.org/wiki/Q116758847","https://g.co/kg/g/11csrrqr2z"]},"Meta",{"@type":"Thing","name":"Meta","sameAs":["https://en.wikipedia.org/wiki/Meta_Platforms","https://www.wikidata.org/wiki/Q225827","https://g.co/kg/m/0hmyfsv"]},"Gemini",{"@type":"Thing","name":"Gemini","sameAs":["https://en.wikipedia.org/wiki/Gemini_(language_model)","https://www.wikidata.org/wiki/Q111082","https://g.co/kg/g/11vclq3pb3"]},"Claude",{"@type":"Thing","name":"Claude","sameAs":["https://en.wikipedia.org/wiki/Claude_(AI)","https://www.wikidata.org/wiki/Q235890","https://g.co/kg/g/11kq5ghr35"]},"Artificial Intelligence Act",{"@type":"Thing","name":"Artificial Intelligence Act","sameAs":["https://en.wikipedia.org/wiki/Artificial_Intelligence_Act","https://www.wikidata.org/wiki/Q108456694","https://g.co/kg/g/11qh3d3r2w"]},"Hugging Face",{"@type":"Corporation","name":"Hugging Face","sameAs":["https://en.wikipedia.org/wiki/Hugging_Face","https://www.wikidata.org/wiki/Q108943604","https://g.co/kg/g/11s0fn67cr"]}],"@id":"https://www.richeyweb.com/blog/development/ai-watermark-detection-is-theater#article","isPartOf":{"@id":"https://www.richeyweb.com/blog/development/ai-watermark-detection-is-theater#webpage"},"publisher":{"@id":"https://www.richeyweb.com/#organization"},"keywords":"AI watermark detection, browser plugin, highlight any block of text, AI watermark, statistical watermarks, secret keys, tokenizer, generation configuration, sampling time, OpenAI-watermarked text, Gemini, Claude, shared detector, third party, Responsible GenAI Toolkit, SynthID-Text, Nature paper, open-source notes, key-dependent bias, detector, ranking guidance, scaled content abuse policy, unoriginal pages, manipulating rankings, manually fact-check, hallucinate, Quality rater guidelines, Lowest quality, detector scores, ranking factors, generative AI content guidance, October 2026 changes, statistical bias, token probabilities, durable signature, rewording, humanizers, multi-pass paraphrasing, translation, ordinary editing, HumanizerBench, detector limits, Open-sourcing, production keys, watermark radioactivity, fine-tuning, watermarked text, residual bias, parameter space, task fine-tuning, Watermarking Makes Language Models Radioactive, unauthorized knowledge distillation, Can LLM Watermarks Robustly Prevent Unauthorized Knowledge Distillation","articleSection":"Development","url":"https://www.richeyweb.com/blog/development/ai-watermark-detection-is-theater"}]}
```