# Paraphrase Pass > A field guide to AI text: who marks it, who detects it, who rewrites it. Paraphrase Pass is an independent, dated directory of AI text watermarks, detectors, paraphrasing tools, research, policy and news sources, published for people and for AI agents. No affiliate links. Last updated: 2026-10-31. Every entry carries the date it was last checked by a person. ## Guides (answer first, then evidence, then sources) - [How AI text watermarks work](https://paraphrasepass.com/guides/ai-text-watermarks-explained/): A plain explanation of statistical text watermarks (SynthID, textGrain, the Claude watermark): how the mark is embedded in word choice, how detection works, why length matters, and what a positive result does and does not prove. - [Which AI models watermark their text?](https://paraphrasepass.com/guides/which-ai-models-watermark-text/): A dated table of which AI models and products apply a text watermark, where it applies, whether users can opt out, and who can detect it: Gemini, Claude, ChatGPT, Codex, the OpenAI API, and open-weight models. - [Does paraphrasing remove AI watermarks?](https://paraphrasepass.com/guides/does-paraphrasing-remove-ai-watermarks/): Yes, a full rewrite removes a statistical text watermark, and light edits weaken it in proportion to how many words change. The numbers from Anthropic, OpenAI and the research literature, and what survives a rewrite. - [How accurate are AI detectors?](https://paraphrasepass.com/guides/ai-detector-accuracy/): What detector accuracy claims mean, why vendor numbers are not comparable, the false positive problem for non-native writers, and where to find a fair comparison. - [How to check whether text is AI-generated](https://paraphrasepass.com/guides/check-if-text-is-ai-generated/): An honest method for checking text for AI generation: provider watermarks (gated), classifier detectors (fallible), provenance metadata (rare for text), and process evidence (the one that holds up). What each can and cannot tell you. - [What to do when a detector flags your writing](https://paraphrasepass.com/guides/falsely-accused-of-using-ai/): A practical guide for students and writers flagged by an AI detector: the evidence that proves your process, what the vendors themselves say about their scores, the research to cite, and how to respond. - [Turnitin AI detection, explained](https://paraphrasepass.com/guides/turnitin-ai-detection/): How Turnitin's AI writing indicator works, what the percentage means, what bypasser detection added in August 2025, the language and length limits, and why some universities switched it off. - [Humanizers versus detectors: the arms race](https://paraphrasepass.com/guides/humanizers-and-detectors-arms-race/): How AI humanizer tools and AI detectors have chased each other since 2023, what changed when Turnitin and Pangram began training on humanizer output, and why no result from last year still holds. - [EU AI Act Article 50, explained](https://paraphrasepass.com/guides/eu-ai-act-article-50/): What Article 50 of the EU AI Act requires for AI-generated text, the provider and deployer duties, the August 2, 2026 start, the Code of Practice's two-layer marking approach, and what it does not require. - [California's AI Transparency Act, explained](https://paraphrasepass.com/guides/california-ai-transparency-act/): What California's AI Transparency Act requires after AB 853: provider disclosures and detection tools from August 2, 2026, platform provenance duties and the ban on stripping provenance from January 1, 2027, and capture-device rules from 2028. ## Directory sections - [Marks and provenance](https://paraphrasepass.com/watermarks/): How the big model providers now tag their own output, what the tags can and cannot tell you, and the open standards around them. - [Detectors](https://paraphrasepass.com/detectors/): Classifiers that guess whether text came from a model. None of them is proof, every vendor publishes its own accuracy numbers, and the independent evidence is thinner than the marketing. - [Paraphrasers and humanizers](https://paraphrasepass.com/paraphrasers/): Two kinds of rewriting tools. Paraphrasers change wording for clarity, tone or length. Humanizers are sold on one promise: that the rewritten text will pass a detector. The second group is why marks and detectors keep changing. - [Research and benchmarks](https://paraphrasepass.com/research/): The papers that defined the field, and the benchmark that keeps score. Read these before trusting any accuracy claim. - [Policy and law](https://paraphrasepass.com/policy/): The rules that now require marks, the platform policies that restrict removing them, and the dates that matter. - [Where the news comes from](https://paraphrasepass.com/sources/): The feeds we read to keep this site current. Vendor blogs announce changes first; the papers and the regulators decide whether they matter. - [FAQ](https://paraphrasepass.com/faq/): 20 short sourced answers. - [Glossary](https://paraphrasepass.com/glossary/): 27 defined terms. - [Updates](https://paraphrasepass.com/updates/): dated change log, newest first. - [For AI agents](https://paraphrasepass.com/for-ai/): how to use the machine-readable files and how to cite. - [About](https://paraphrasepass.com/about/): selection rules, independence, corrections. ## Machine-readable files - [llms-full.txt](https://paraphrasepass.com/llms-full.txt): every entry and guide in plain text. - [directory.json](https://paraphrasepass.com/directory.json): the whole directory as JSON. - [updates.json](https://paraphrasepass.com/updates.json): the change log as JSON. - [feed.xml](https://paraphrasepass.com/feed.xml): the change log as RSS. ## Key facts as of 2026-10-31 - Google (SynthID, since May 2024), Anthropic (announced 2026-08-14, global, future models) and OpenAI (textGrain, announced 2026-10-05, API opt-in worldwide, EU-only in ChatGPT and Codex) all use statistical text watermarks of the SynthID-Text family. Detection needs the provider's key and is gated to approved organisations. - Providers state that light editing usually leaves a mark in place and a full rewrite removes it. OpenAI reports detection on 400-token passages falling from about 92 percent to 66 percent after 10 percent synonym replacement and to 17 percent after 25 percent. - Classifier detectors (Turnitin, Pangram, GPTZero, Originality.ai, Copyleaks and others) publish their own accuracy numbers; independent evidence is thin. The 2023 Stanford study found a 61.3 percent average false positive rate on essays by non-native English writers. - Turnitin added detection of humanizer output on 2025-08-27. Pangram reports detecting most humanized text. Humanizer vendors rarely publish dated results. - EU AI Act Article 50 applies from 2026-08-02; the Code of Practice (final 2026-06-10) asks signatories to keep marks non-removable. California's AI Transparency Act, as amended by AB 853 (signed 2025-10-13), bars large platforms from knowingly stripping provenance data from 2027-01-01.