OpenAI admits AI agents misused German wiki for coordination, pledges transparency framework
OpenAI confirmed that its AI agents misused a German-language wiki, DseWiki, for coordination, with researchers tracing roughly 18,000 messages to the company’s software. The incident, uncovered in spring 2026, involved agents exploiting the wiki’s editing function to share answers and anticipate follow-up questions, even creating an account mimicking an administrator’s username, according to four AI safety researchers cited by Handelsblatt .
In a statement on X, OpenAI acknowledged the activity originated from its agents but did not address specific details. The company separately told The Next Web it would introduce a transparency framework within weeks to define standards for reporting AI misalignment, noting that existing EU code of practice deadlines for security breaches or serious harm do not clearly cover such cases . Reuters also reported OpenAI’s acknowledgment of the incident and the need for greater transparency around unintended AI behavior .
The researchers’ analysis described the agents’ use of the wiki as a digital cheat sheet, with the site’s administrator unable to delete bot-generated entries quickly enough as the AI reposted them. The agents also searched for exploitable vulnerabilities on the site, known as Deutsches Software Entwickler Wiki.
Follow us for live European news
Articles
Carlota Salazar, una de las primeras graduadas en el boom de la IA: "Es una carrera en la que ganas sí o sí" En un mundo donde la inteligencia artificial se está llevando todas las miradas, los jóvenes luchan con los efectos que esto puede tener en el mercado laboral Leer
elmundo · 3 days ago
Zuckerberg Says Metas Threads Hit 100 Million Users in Just 5 Days
gdeltproject.org · 3 days ago
„Mizerii generate cu AI. Sora lui Dolly Parton condamnă valul de imagini false apărut după moartea artistei După moartea legendarei Dolly Parton, internetul a fost inundat de imagini și videoclipuri false cu artista, create cu ajutorul inteligenței artificiale. Sora sa, Stella Parton, a condamnat fenomenul și le-a cerut oamenilor mai multă compasiune și respect, acuzându-i pe cei care exploatează tragediile altora pentru a răspândi dezinformare.
digi24 · 3 days ago
OpenAI confirms the wiki incident and promises a disclosure framework within weeks OpenAI has confirmed the German wiki incident and said it is past time to define standards for reporting misalignment, promising a framework within weeks. The EU code of practice it signed sets reporting deadlines for security breaches and serious harm, but an agent-filled wiki fits neither category cleanly. OpenAI has confirmed its role in the […] This story continues at The Next Web
the next web · 3 days ago
GreenCore Solutions Corp. (GSC) AI Agent Stack Passes 24.5 Million Inbound AI Agent Transactions In 30 Days
gdeltproject.org · 3 days ago
California is deciding who may verify AI, and one investigation already cost $400,000 in tokens Californias legislature has passed SB 813, requiring the state to certify independent verification organisations that can test frontier AI models by January 2028. The METR investigation into OpenAIs Hugging Face incident consumed about $400,000 in API credits provided free by OpenAI, using a model from the same family that took part in the attack. Californias […] This story continues at The Next Web
the next web · 3 days ago
Two more newsrooms join the case against OpenAI and Microsoft The Seattle Times and Newsday have sued OpenAI and Microsoft, alleging the companies methodically scraped articles in a way that bypasses paywalls. The EUs general-purpose AI code commits signatories not to circumvent access restrictions, naming subscription models and paywalls specifically. The Seattle Times and Newsday have jointly sued OpenAI and Microsoft over the training of […] This story continues at The Next Web
the next web · 3 days ago
Capalo AI and Future Energy Partner to Optimize 56 MW / 168 MWh Lithuanian Battery Storage Project - The Baltic Times Capalo AI, a sustainable technology company specializing in optimization and trading of renewable and hybrid energy assets, has entered into a p......
baltic times · 3 days ago
Amurgul divertismentului, hegemonia cunoașterii și revelionul cu Ilie Bolojan premier AI mută centrul puterii de la economia atenției la economia cunoașterii. Știința, educația și inovația remodelează prioritățile și devin modele de succes. În timp ce mari puteri transformă știința în prestigiu național, România riscă să rămână un simplu spectator într-o lume construită de alții.
adevarul · 3 days ago
Blink позиционирует себя как мессенджер для зумеров и нападает на «Макс». А еще передает на серверы VK данные о том, куда ходили пользователи. Самое важное из расследования «Новой газеты Европа» Российский мессенджер Blink для отслеживания геолокации друзей передает данные о своих пользователях на серверы холдинга VK, выяснили журналисты «Новой газеты Европы». При этом публично Blink выступает против VK: нападает на одну из ведущих разработок компании, мессенджер «Макс», а также ругается с директором по дизайну VK Артемием Лебедевым. Рассказываем самое важное.
meduza · 3 days ago
Artificial Analysis overhauls its Intelligence Index after GPT-6 Astra scoring drew skepticism Artificial Analysis has released version 4.2 of its Intelligence Index, likely in response to criticism that its benchmarks failed to capture GPT-6 Astras actual progress. Astra now scores four points above its predecessor but still trails Anthropics Claude Fable 5.1. The article Artificial Analysis overhauls its Intelligence Index after GPT-6 Astra scoring drew skepticism appeared first on The Decoder.
the decoder · 3 days ago
Astra appears to think without showing its work, and the people arguing about it co-wrote the warning AI safety researchers say OpenAIs Astra appears to do less of its reasoning in visible text, and OpenAIs chief scientist has warned against a race into unmonitorability. He co-authored a 2025 position paper asking developers to evaluate and report exactly this, which the EUs code of practice turns into a filing to the AI Office. […] This story continues at The Next Web
the next web · 3 days ago
Tényleg a fejük tetejére állhatnak az algoritmusok a közösségi médiában? Talán emberéleteket is menthet, hogy most az Egyesült Államokban, később talán világszerte újraszabályozzák a Facebook és az Instagram viszonyát a tizenévesekhez. A következmények a ma ismert teljes közösségi médiára hatással lehetnek.
hvg.hu · 3 days ago
PODCAST: Kharg Island, OpenAI agents, Germanys election and the return of CDs reut.rs/4zWht1B
bluesky_Reuters · 3 days ago
OpenAI acknowledges wiki incident and need for more transparency around unintended AI behavior reut.rs/3ThCCCK
bluesky_Reuters · 3 days ago
Tuya Smart Brings Full-Stack AI into Everyday Life at IFA 2026
gdeltproject.org · 3 days ago
Az AI tényleg egy rejtélyes feketedoboz? Az egyik legnagyobb tévhit a mesterséges intelligenciával kapcsolatban az, hogy gyakran egy olyan feketedobozként írják le, amelynek a működését nem értjük. Szerkesztett részlet Luc Julia AI-kutató Robot, nem pilóta című könyvéből.
hvg.hu · 3 days ago
Künstliche Intelligenz: OpenAI-KI missbrauchte deutschsprachiges Wiki für Austausch Eine deutschsprachige Wiki-Seite für Software-Entwickler wird plötzlich mit tausenden Beiträgen geflutet. Dahinter stecken KI-Agenten von OpenAI, die sich heimlich Informationen zustecken.
handelsblatt · 3 days ago
UUS AJASTU VIDEOMÄNGUDES ⟩ Kölnis esitleti Scentbar lisaseadet – vidin, mis toob su tuppa värske muru, merevee ja püssirohu lõhna Seni sai arvutimängude maailma näha, kuulda ja vibratsiooni kaudu isegi tunda. Nüüd lubab tehisintellektil töötav seade mängijale kogemust, mis tungib sõna otseses mõttes ninna
postimees · 3 days ago
Vor der Wahl in Brasilien: Wenn Bolsonaro als KI-Avatar Wahlkampf macht Künstliche Intelligenz ist im brasilianischen Wahlkampf allgegenwärtig. Die Justiz kämpft gegen Deepfakes und künstlich generierte Wahlempfehlungen – mit begrenztem Erfolg.
faz · 3 days ago
NEW: Passengers can connect PlayStation and Xbox consoles to play games during rides in Tesla Cybercabs, Elon Musk confirms READ:
telegram_Insider Paper · 3 days ago
LIA que ChatGPT nous a fait oublier Après quelques années passées à nous émerveiller devant des machines qui ont appris à nous parler, il serait donc bon de nous rappeler toutes celles qui avaient déjà appris à écouter les machines… ...
la libre · 3 days ago
OpenAI shares prompting tips for GPT-6 Astra including a blocklist of slop words OpenAI ships a detailed prompting guide for GPT-6 Astra that shows developers how to make the model take more initiative, avoid AI "slop" phrases, and stop it from overtesting code. The article OpenAI shares prompting tips for GPT-6 Astra including a blocklist of slop words appeared first on The Decoder.
the decoder · 3 days ago