All articles

Reading paths

1. 01

   New to watermarking

   Understand what the phrase even means before forming an opinion.

   1. What Is AI Watermarking?
   2. →AI Text Watermark
   3. →AI Watermark vs AI Detector
2. 02

   Claude user

   Work out what is actually true about Claude's mark, and what you can check.

   1. Claude Watermark
   2. →Hidden Unicode AI Watermark
   3. →Remove Claude Watermark
   4. →Claude Watermark Detector
3. 03

   Policy and compliance

   Separate the provider marking duty from the deployer disclosure duty.

   1. EU AI Act and AI Watermarking
   2. →Machine-Readable AI Marking
   3. →AI-Generated Content Disclosure
4. 04

   Researcher

   Get to the mechanism and the attack literature without the explainer layer.

   1. Token Probability Watermarking
   2. →Statistical Text Watermarking
   3. →Text Watermark Robustness
   4. →AI Watermark Spoofing
5. 05

   Verifying a file

   Find out whether a specific image, video, or audio file carries provenance data.

   1. C2PA Content Credentials for AI Provenance
   2. →Metadata AI Watermarking
   3. →C2PA vs SynthID

Browse everything

43 of 43 shown

• AI Text Watermark Removal

  Original dataResearch/proposal

  Before you pay for a remover: the under $50 lab attack that did work, the signal loss found inside SynthID Text, and the EU exemptions few ever cite.

  Core

• AI Text Watermark: How Statistical Watermarking Works

  Research/proposal

  No hidden characters to find. A statistical watermark rigs which words the model picks. The mechanism, the one concrete number, and where it breaks.

  Core

• What Is AI Watermarking?

  Confirmed

  Visible labels, invisible signals, C2PA metadata, statistical watermarking: four mechanisms hiding behind one name, and how each one actually works.

  Core

• AI Watermark Removal: What It Can and Cannot Mean

  Research/proposal

  Metadata strips. Invisible characters clean off. The statistical watermark is another story. What removal really means, and the $50 attack that broke one.

  Core

• Can AI Watermarks Be Removed?

  Research/proposal

  Metadata, pixels, audio, and text all fail differently. The signal-by-signal answer, and where the evidence flat-out contradicts the marketing.

  Core

• AI Watermark vs AI Detector

  Confirmed

  GPTZero, Originality.ai, and ZeroGPT don't check for watermarks at all. Why a planted signal and an after-the-fact guess disagree, with the evidence.

  Core

• Claude Watermark: Confirmed Facts, Reports, and Rumors

  Verified 2026-09-25Confirmed

  Opus 5 carries Claude's watermark from Sep 14, 2026, and Fable 5, Sonnet 5 and Opus 4.8 are dated to Sep 30. Model by model, with Anthropic's sources.

  Provider trackers

• Anthropic Watermark: The Company, the Law, the Timeline

  Verified 2026-09-25Confirmed

  Anthropic didn't pick August 2026, EU AI Act Article 50 did. The commitments, the C2PA stance, and the full dated timeline behind the rollout.

  Provider trackers

• ChatGPT Watermark: Is Anything Actually Watermarked?

  Verified 2026-09-03Confirmed

  What the ChatGPT watermark is, what OpenAI shelved, and what removal can mean. Images and audio carry real provenance signals; ordinary chat text carries none.

  Provider trackers

• Gemini Watermark: Is Gemini Watermarked, and Where?

  Verified 2026-09-03Official announcement

  Is Gemini watermarked? SynthID is confirmed on images, audio and video. For text, Google's own pages contradict each other, and this is where they split.

  Provider trackers

• Is Grok Watermarked? Images and Video Yes, Text Not Documented

  Verified 2026-08-13Confirmed

  xAI puts a visible watermark on Grok images and video and documents nothing about text. What's confirmed, what's contested, and what's just marketing.

  Provider trackers

• Is Llama Watermarked? Meta's Own Documents Say Nothing

  Rumor/speculation

  Meta marks its images and audio, yet Llama's model cards never mention text watermarking. Even Meta's EU paperwork skips text. The silence is the answer.

  Provider trackers

• DeepSeek Watermark: Is DeepSeek Output Watermarked?

  Verified 2026-09-10Confirmed

  Is DeepSeek watermarked? Chinese law forces visible AI labels in three places. No DeepSeek document describes a hidden mark, and its terms forbid removing them.

  Provider trackers

• Meta AI Watermark: Content Seal, Audio, and Shelved Research

  Verified 2026-08-12Confirmed

  Meta ships two watermarking product lines and open-sourced a third for text. So why does Meta AI chat text carry no mark today? The full map, verified.

  Provider trackers

• LLM Watermarking: Claude vs Gemini vs ChatGPT vs Llama

  Verified 2026-09-25Research/proposal

  Claude, Gemini, ChatGPT, Grok, Llama, and DeepSeek compared on text watermarking in 2026, plus the real reason OpenAI shelved its own system.

  Provider trackers

• Remove Claude Watermark: Real Mark, No Way to Check Yet

  Original dataVerified 2026-09-25Confirmed

  Anthropic's detector shipped in Sep 2026, to regulators and newsrooms only. You still cannot check your own text, so no remover can prove it worked.

  Removal guides

• Remove Gemini Watermark: What SynthID Text Can Survive

  Research/proposal

  Google says SynthID survives light edits but not a full rewrite. One forum reply even disputes the API applies it. Read the fine print before you act.

  Removal guides

• Token Probability Watermarking

  Verified 2026-08-15Research/proposal

  Token probability watermarking biases word choice mid-generation. The patent describing SynthID without naming it, and the $50 attack that stole its rule.

  How the mechanism actually works

• Statistical Text Watermarking

  Verified 2026-08-15Research/proposal

  Statistical text watermarking adds nothing to your text: a secret key steers word choice instead. How the scheme works, and how researchers cracked one.

  How the mechanism actually works

• Text Watermark Robustness: What Actually Survives an Edit

  Research/proposal

  Text watermark robustness has a floor: detection wants roughly 800 tokens. What survives an edit, and why four SynthID Text audits tell a rockier story.

  How the mechanism actually works

• Paraphrasing AI Watermarks: What One Rewrite Actually Does

  Research/proposal

  Paraphrasing is the most-studied attack on AI text watermarks. The 2023 paper that said it wins, and 2026 data showing 98.3% signal loss from one rewrite.

  How the mechanism actually works

• Hidden Unicode AI Watermark: Proves Almost Nothing

  Original dataVerified 2026-08-15Community discussion

  Found an invisible character in Claude or ChatGPT text? Four Unicode characters explain nearly all of them, and none proves AI wrote it. See which four.

  How the mechanism actually works

• Zero-Width Space Watermark: Real Technique, Wrong Mechanism

  Original dataVerified 2026-08-15Community discussion

  Zero-width space steganography is real and decades old. It is also not how Claude or Gemini watermark text. We scanned 96 Claude outputs and found 0.

  How the mechanism actually works

• SynthID Watermark: Text, Images, Audio, and Video

  Confirmed

  How SynthID hides its mark in images, audio and video, how SynthID Text steers token sampling as the model writes, and what red teams found attacking it.

  How the mechanism actually works

• AI Watermarking Research: A Map of the Literature

  Research/proposal

  The whole field on one page: the 2023 green-list paper, the schemes built on it, the under $50 attack that steals watermarks, and SynthID's patents.

  How the mechanism actually works

• AI Watermark Spoofing: Forging Someone Else's Signal

  Research/proposal

  Spoofing forges a model's watermark onto text it never wrote. One lab pulled it off for under $50, then published the test that catches the forgery.

  How the mechanism actually works

• Do Em Dashes Mean AI Wrote It? What the Data Shows

  Original dataCommunity discussion

  Two of six chatbots wrote zero em dashes on identical prompts. Where the rule came from, and why it fails, before you accuse a human writer.

  How the mechanism actually works

• AI Text Watermark Detector

  Verified 2026-09-25Research/proposal

  How AI text watermark detectors actually work, why Claude has no public one, why Google's checker may verify nothing, and where GPTZero really fits.

  Detection

• Claude Watermark Detector: Private Preview, Not Public

  Original dataVerified 2026-09-25Confirmed

  Anthropic's detection API shipped on 1 Sep 2026, to regulators, media, fact-checkers and researchers only. What that leaves you, and what to check instead.

  Detection

• AI Watermark Detector

  Confirmed

  AI watermark detectors scan image, video, and audio, yet one scheme mismatch turns real AI content into a false negative. A clean result proves nothing.

  Detection

• AI Detector False Positives: Why They Happen So Often

  Confirmed

  AI detector false positives are documented, common, and mechanical. Why honest writing gets flagged, and why a watermark check beats any detector score.

  Detection

• EU AI Act and AI Watermarking: Article 50 Explained

  Verified 2026-09-05Official announcement

  EU AI Act Article 50 now makes marking AI content the law. Who is covered, what counts as compliance, the deadline, and what noncompliance costs.

  Regulation and disclosure

• Machine-Readable AI Marking: EU AI Act Article 50 Rules

  Verified 2026-08-13Official announcement

  Article 50 makes machine-readable AI marking law without naming a technology. What it demands, who it binds, and why EU enforcement is so uneven.

  Regulation and disclosure

• AI-Generated Content Disclosure: EU Article 50 Duties

  Official announcement

  Article 50's disclosure duty is not the marking duty, and it lands on different people. Who must disclose AI-generated content, and who is exempt.

  Regulation and disclosure

• C2PA Content Credentials for AI Provenance

  Confirmed

  Inside a C2PA Content Credentials manifest: what it records, how hard binding makes tampering visible, and what a verified credential still can't prove.

  Provenance and metadata standards

• C2PA vs SynthID

  Verified 2026-08-15Confirmed

  C2PA credentials and SynthID watermarks fail in opposite ways. See which one survives a screenshot, and who is actually allowed to run each detector.

  Provenance and metadata standards

• AI Content Provenance: What It Proves, and What It Doesn't

  Confirmed

  What AI content provenance really records, how it differs from watermarking and AI detection, and the limits the vendors admit in their own words.

  Provenance and metadata standards

• AI Signature: Watermarks, Metadata, and Style Signals

  Community discussion

  Four things get called an AI signature: a signed C2PA manifest, an invisible watermark, a detector's verdict, and a writing tic like the em dash.

  Provenance and metadata standards

• Metadata AI Watermarking

  Confirmed

  What a C2PA manifest quietly stores about your file, the two free tools that read and delete one, and why metadata is not a hidden watermark at all.

  Provenance and metadata standards

• Image AI Watermarking: OpenAI, Google, Meta, and Amazon

  Confirmed

  OpenAI, Google, Meta and Amazon each hide a mark in AI images. See what the invisible layer survives and the exact point where tests saw it fade.

  Other media (image, video, audio)

• Video AI Watermarking: Sora, SynthID, and Nova Reel

  Confirmed

  Sora, SynthID and Nova Reel all watermark AI video, each differently. One test even found OpenAI's own provenance claim backwards. Here's who marks what.

  Other media (image, video, audio)

• Audio AI Watermarking: OpenAI, Meta, and Google SynthID

  Confirmed

  OpenAI shipped its audio watermark two days before an EU deadline. Meta did it a year earlier, and its patent names Ian Goodfellow. How each one works.

  Other media (image, video, audio)

• Community AI Watermark Questions and Rumors

  Community discussion

  What Reddit, Hacker News, and X really believe about AI watermarks, including the viral SynthID crack claim a correction cut down to just 16 percent.

  Community