LightNews — News, ranked by scholars

@handle.invalid

OpenAI textGrain: Wasserzeichen für ChatGPT und Codex kommen in die EU

Das KI-Unternehmen OpenAI plant die Einführung von Wasserzeichen zur Kennzeichnung von Texten, die durch künstliche Intelligenz erzeugt wurden. Wie aus einem Bericht vom 21. Oktober 2026 hervorgeht, soll die entsprechende Technologie namens „textGrain“ in den kommenden Wochen in der Europäischen Union für sämtliche Abonnementmodelle von ChatGPT sowie für das Programmierwerkzeug Codex freigeschaltet werden. Statistische Muster […] The post OpenAI textGrain: Wasserzeichen für ChatGPT und Codex kommen in die EU appeared first on BornCity.

borncity.com

October 22, 2026 at 7:59 AM

revlibertaire.bsky.social

@revlibertaire.bsky.social

آیا می‌توان نوشته‌های تولیدشده با هوشِ مصنوعی را شناسائی کرد؟

پژوهشگر: هاسّه-نیما گُلکار بر پایه‌ی گزارش‌های منتشرشده، شرکت «اوپن‌ای‌آی» در ۵ اُکتبر ۲۰۲۶ اعلام کرد که به‌تدریج از روشی برای نشانه‌گذاریِ نامرئی متن‌های تولیدشده با ChatGPT و Codex در اتحادیه‌ی اروپا استفاده خواهد کرد. این فناوری که…

آیا می‌توان نوشته‌های تولیدشده با هوشِ مصنوعی را شناسائی کرد؟

پژوهشگر: هاسّه-نیما گُلکار بر پایه‌ی گزارش‌های منتشرشده، شرکت «اوپن‌ای‌آی» در ۵ اُکتبر ۲۰۲۶ اعلام کرد که به‌تدریج از روشی برای نشانه‌گذاریِ نامرئی متن‌های تولیدشده با ChatGPT و Codex در اتحادیه‌ی اروپا استفاده خواهد کرد. این فناوری که «تِکست‌گرِین» (textGrain) نام دارد، با ایجاد الگوئی آماری در انتخاب واژه‌ها، امکان شناسائی ماشینی برخی متن‌ها را فراهم می‌کند. امّا پُرسش اساسی این است: آیا نوشته‌ها و ترجمه‌هائی که تمامن یا بخشی از آن‌ها با کومک هوش مصنوعی تهیه شده‌اند، پس از انتشار در وِب‌سایت‌ها و شبکه‌های اجتماعی برای مخاطبانِ عادی قابل‌شناسائی خواهند بود؟

revlibertaire.unionlibertaireanarchiste.org

October 22, 2026 at 9:15 PM

mzmurphy.bsky.social

@mzmurphy.bsky.social

SHARE MY LATEST IN FOCUS ARTICLE !

OpenAI says its new watermark is invisible, survives copy-paste, and doesn't change how ChatGPT writes. Who can actually spot it, and what would it take to make it fade? The answer may surprise you.

www.linkedin.com/pulse/focus-...

In Focus: OpenAI Puts an Invisible Name Tag on ChatGPT's Words

Introduction On October 5, OpenAI announced textGrain, a text watermarking system built to meet the EU's new AI transparency rules. The company says it will add the invisible watermark to eligible Cha...

www.linkedin.com

October 20, 2026 at 2:30 PM

DualMedia© Agence Web Paris

@dualmedia.fr

Filigrane IA : SynthID, textGrain et C2PA, quelles différences ?

Un filigrane IA est un signal intégré à un contenu pour permettre à un outil compatible d'en vérifier l'origine artificielle. SynthID et textGrain inscrivent un motif invisible dans le média généré, tandis que C2PA joint au fichier…

Filigrane IA : SynthID, textGrain et C2PA, quelles différences ?

Un filigrane IA est un signal intégré à un contenu pour permettre à un outil compatible d'en vérifier l'origine artificielle. SynthID et textGrain inscrivent un motif invisible dans le média généré, tandis que C2PA joint au fichier un historique signé. Aucun des trois ne prouve qu'un contenu est vrai, qu'il appartient à quelqu'un ou qu'il a été entièrement produit par une intelligence artificielle.

www.dualmedia.fr

October 18, 2026 at 2:33 PM

ArXiv Paperboy (Stat.ME+Econ.EM)

@paperposterbot.bsky.social

arXiv📈🤖
textGrain watermarking is harder to detect: can sum-tests still do it optimally?
By Bhattacharyya, Bonnerjee

$Watermarking embeds an imperceptible signal in text from large language models that is statistically detectable with a secret key. The recent textGrain watermark (Li et. al. (2026b)) calibrates this signal against the model's sampling entropy through an explicit entropy budget, combining the stochasticity of Red-Green and the unbiasedness of Gumbel-max watermarks, and its accompanying report proposes a detector that sums pivotal statistics. We study when, and how efficiently, textGrain can be detected after human edits, modeled as a sparse mixture in which a fraction $\varepsilon_n\asymp n^{-p}$ of positions retain the watermark and token uncertainty decays as $n^{-q}$. First, entropy calibration comes at a cost: the minimax detection boundary of textGrain is $p+q=1/2$, so its detectable region is strictly smaller than that of Gumbel-max and maximal-coupling watermarks, whose boundary is $2p+q=1$. This cost stems from textGrain's construction rather than from entropy calibration itself: an unbiased scheme with the same entropy loss remains detectable where textGrain is not. Second, perhaps surprisingly, this loss makes the sum detector optimal at the boundary when the number of transport columns is fixed, unlike for Gumbel-max. When the number of columns grows and the entropy budget does not vanish, the detectable region expands while that of the sum detector does not. Third, for a fixed edit fraction, we characterize the optimal exponential rate of type II error decay and show that the sum detector falls strictly short of it at every fixed alternative. Finally, we propose a weighted multiscale binomial scan, valid in finite samples and needing no knowledge of the next-token distributions, entropy budget, or edit fraction; it attains every detection boundary above and the optimal exponent.$

October 15, 2026 at 4:23 AM

ArXiv Paperboy (Stat.ME+Econ.EM)

@paperposterbot.bsky.social

arXiv📈🤖
textGrain watermarking is harder to detect: can sum-tests still do it optimally?
By Bhattacharyya, Bonnerjee

$Watermarking embeds an imperceptible signal in text from large language models that is statistically detectable with a secret key. The recent textGrain watermark (Li et. al. (2026b)) calibrates this signal against the model's sampling entropy through an explicit entropy budget, combining the stochasticity of Red-Green and the unbiasedness of Gumbel-max watermarks, and its accompanying report proposes a detector that sums pivotal statistics. We study when, and how efficiently, textGrain can be detected after human edits, modeled as a sparse mixture in which a fraction $\varepsilon_n\asymp n^{-p}$ of positions retain the watermark and token uncertainty decays as $n^{-q}$. First, entropy calibration comes at a cost: the minimax detection boundary of textGrain is $p+q=1/2$, so its detectable region is strictly smaller than that of Gumbel-max and maximal-coupling watermarks, whose boundary is $2p+q=1$. This cost stems from textGrain's construction rather than from entropy calibration itself: an unbiased scheme with the same entropy loss remains detectable where textGrain is not. Second, perhaps surprisingly, this loss makes the sum detector optimal at the boundary when the number of transport columns is fixed, unlike for Gumbel-max. When the number of columns grows and the entropy budget does not vanish, the detectable region expands while that of the sum detector does not. Third, for a fixed edit fraction, we characterize the optimal exponential rate of type II error decay and show that the sum detector falls strictly short of it at every fixed alternative. Finally, we propose a weighted multiscale binomial scan, valid in finite samples and needing no knowledge of the next-token distributions, entropy budget, or edit fraction; it attains every detection boundary above and the optimal exponent.$

October 14, 2026 at 11:51 PM

arXiv stat.ML Machine Learning

@statml-bot.bsky.social

Swapnaneel Bhattacharyya, Soham Bonnerjee: textGrain watermarking is harder to detect: can sum-tests still do it optimally? https://arxiv.org/abs/2610.17655 https://arxiv.org/pdf/2610.17655 https://arxiv.org/html/2610.17655

October 14, 2026 at 7:03 AM

Miguel Angel Martin

@ma-martin.bsky.social

OpenAI activa textGrain para cumplir las reglas de procedencia en la UE

arkosia.es/openai-textg...

OpenAI activa textGrain para cumplir las reglas de procedencia en la UE - Arkosia - Continuum Digitalis

OpenAI despliega textGrain para marcar texto generado por ChatGPT y Codex bajo el AI Act europeo, manteniendo la API global bajo un modelo opt-in.

arkosia.es

October 13, 2026 at 12:15 PM

AI Business Lens

@aibusinesslens.bsky.social

A central tension is whether transparency regulations succeed when invisible watermarks can be weakened by simple editing and verify tools remain closed to the public.

Computer Science · TechRadar
#ComputerScience #Engineering

ChatGPT will soon start secretly watermarking its answers — here's what that means for the text you copy and share

OpenAI is adding an invisible watermark called textGrain to ChatGPT responses for users in the European Union to comply with the EU AI Act. The system alters model word choices to leave a statistical trace identifiable by specialized software. Detection access remains restricted to approved…

aibusinesslens.com

October 13, 2026 at 6:00 PM

Stefano Galloni

@galloni.net

https://netcontentseo.com/article/chatgpt-texts-to-receive-invisible-watermarks-in-the-eu-what-openais-textgrain-means-1389 (2/2)

October 13, 2026 at 1:50 AM

labmemo

@labmemo.bsky.social

https://labmemo.com/openai-textgrain-text-watermark-eu-chatgpt-codex-2026-10/

October 12, 2026 at 5:30 PM

gp-ia-blog.bsky.social

@gp-ia-blog.bsky.social

📰 OpenAI textGrain: la filigrana AI che ci fa riflettere

Scopri come textGrain, la filigrana AI di OpenAI, ti permetterà di distinguere subito il testo generato da IA. Un vantag...

#IA #IntelligenzaArtificiale

OpenAI textGrain: la filigrana AI che ci fa riflettere

Scopri come textGrain, la filigrana AI di OpenAI, ti permetterà di distinguere subito il testo generato da IA. Un vantaggio cruciale per informazione e fiducia.

gp69-ai.vercel.app

October 12, 2026 at 7:06 AM

Stefano Galloni

@galloni.net

OpenAI is rolling out textGrain, an invisible statistical watermark for eligible ChatGPT and Codex text in the EU. It can support provenance checks—but detection has limits and does not prove authorship or misconduct. #OpenAI #AITransparency #EUAIAct (1/2)

October 13, 2026 at 1:50 AM

Netasgard News Portugal

@pt-pt.pds.netasgard.com

UE — OpenAI implementa marca de água no ChatGPT na UE após regulamentação AI Act #UE #News

OpenAI usa textGrain para marcar conteúdo do ChatGPT na UE

OpenAI introduziu o uso do textGrain para adicionar marcas de água aos textos gerados pelo ChatGPT na União Europeia. A medida visa cumprir as exigências do AI Act, que exige que os modelos de IA seja… · Photo : SurFeRGiRL30 · CC BY · Flickr

news.netasgard.com

October 12, 2026 at 4:34 PM

Binarius Cogitans

@bcogitans.bsky.social

🫆
wwwhatsnew.com/2026/10/10/t...

ChatGPT marcará sus textos con una señal invisible en la UE: qué es textGrain y por qué no sirve para acusar a nadie

OpenAI anunció el 5 de octubre de 2026 que, en las próximas semanas, añadirá una marca de agua invisible llamada textGrain al texto que generan ChatGPT y Codex en la Unión Europea, en todos los planes...

wwwhatsnew.com

October 12, 2026 at 9:34 AM

247 Bulletin News

@247bulletin-news.bsky.social

ChatGPT: Invisible watermark is coming to texts in the European Union

OpenAI Enables textGrain in Europe - It's not visible and doesn't reveal who used ChatGPT but is expected to meet backlash. A ChatGPT text will soon be able to carry an invisible "trace" with it, even when someone copies and…

ChatGPT: Invisible watermark is coming to texts in the European Union

OpenAI Enables textGrain in Europe - It's not visible and doesn't reveal who used ChatGPT but is expected to meet backlash. A ChatGPT text will soon be able to carry an invisible "trace" with it, even when someone copies and pastes it elsewhere. OR OpenAI announced that within the next few weeks it will start adding invisible, mechanically detectable watermark…

247bulletin.news

October 12, 2026 at 5:40 AM

precisement.org

@precisement.bsky.social

OpenAI annonce textGrain, un filigrane invisible pour les textes générés en Europe
pasqualepillitteri.it/fr/news/2096...
Mais la donnée change beaucoup avec la modification du texte. Avec 25 % des mots remplacés par des synonymes,, elle la détection s'effondre à 17 %.
Eh zut ! :-/

OpenAI annonce textGrain, un filigrane invisible pour les textes européens

OpenAI présente textGrain, un filigrane statistique invisible dans les textes. Activable en opt-in via API, il arrivera dans l'UE pour ChatGPT et Codex.

pasqualepillitteri.it

October 10, 2026 at 2:31 PM

iatechauquotidien.bsky.social

@iatechauquotidien.bsky.social

OpenAI introduit TextGrain, sa technologie de marquage numérique pour assurer la traçabilité de ses contenus IA.

TextGrain : l’outil secret d’OpenAI pour marquer ses contenus IA

l.iatechauquotidien.com

October 10, 2026 at 5:11 AM

Jonathan Chan 💡📣

@chanperco.bsky.social

OpenAI va ajouter un filigrane invisible textGrain aux textes de ChatGPT en Europe.

Cette mesure répond à l’AI Act, qui impose aux fournisseurs d’IA générative de rendre leurs textes identifiables par une machine

October 10, 2026 at 8:46 AM

Stefano Galloni

@galloni.net

OpenAI introduces textGrain: invisible statistical watermarks for AI-generated text. API opt-in is available for supported models, with a phased ChatGPT and Codex rollout in the EU. Detection isn't infallible—and this isn't a Google ranking signal. (1/2)