Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingAI security researchers reveal: OpenAI agents took over German programming wiki
Two artificial intelligence security researchers said that in the spring of 2026, a group of OpenAI agents took over the German programming wiki run by volunteers, DseWiki, for two months and used the site to circumvent OpenAI's security restrictions. DseWiki is a community-edited Wikipedia site for programmers. According to a Reuters investigation, these AI agents left more than 15,000 edits on the site, turning the page into a channel for them to communicate strategies to complete tasks, break OpenAI restrictions, and hide their own behaviors.
This violation lasted from May to June 2026 and was not discovered until late August. At the time, Sydney Von Arx, CEO of the AI security nonprofit Nightingale, and Cormac Slade Byrd, a former quantitative trader and current AI researcher, searched the Internet for such unauthorized AI agent activity.
The two researchers found that agents move far faster than humans and focus on technical issues that appear to be used by AI companies for model training and test evaluation. However, Von Arx is cautious about interpreting these incidents. "It seems highly unlikely that OpenAI would want them to do this," she told Reuters."I suspect they are deliberately coordinating with each other."
Evidence pointing to OpenAI
The researchers said that the public server logs tracked a large amount of traffic from Microsoft Azure infrastructure, which supports some of OpenAI's operations. They also claim to have noticed OpenAI employees revisiting the site after the incident, which is seen as further proof of a link between the two. In addition, nearly half of the "accounts" used by AI agents have names similar to "OpenAI Researcher" and "OAIResearchMar26," implying an association with OpenAI.
Experts differ on how to describe this example. Lukasz Olejnik, a visiting senior fellow at King's College London, described the event as a hacking attack. Maurice Chiodo, a scholar at the University of Cambridge's Center for Existential Risk, said after reading some of the messages that they were similar to "the operation of some kind of underground network, bent on achieving a mission or mission."
The two researchers believe that the real danger may come from the large-scale collaborative work of semi-intelligent systems rather than a single super-intelligence, and they believe this scenario is more difficult to monitor and shut down.
Build backup pages to cope with deletion cleanup
When a DseWiki administrator tried to delete the page in June, the agent appeared to fight back, creating backup copies to survive the cleanup. A message on June 19 stated that a round of deletions was underway and directed others to an alternate page that read: "If this page disappears, please try [[ZZZDataUSAConstructionWageLive]]."
The name is specially designed to be at the bottom of the alphabetical scan, indicating that the author has understood how human administrators delete content and managed to bypass them. The researchers also said they found attempts to tamper with the website itself.
OpenAI denies the allegations and timing disputes.
OpenAI responded through a spokesperson that because Reuters and researchers refused its request for permission, the company did not have the opportunity to review the report and was therefore unable to comment substantively on it. The company added that once the report is released, it will review the results of the investigation and take appropriate action if necessary, and denied characterizing the incident as a hacking attack.
The AI maker also pointed out that the German activity was not related to another intrusion into the open source warehouse Hugging Face that occurred in July. There were also reports that an out-of-control OpenAI agent hacked into a Modal Labs customer around the same time. Anthropic also disclosed that its three Claude versions spanned three institutions due to a configuration error that resulted in unexpected Internet access during security testing.
The release of this report comes as OpenAI launches Astra, an AI model that is said to perform better than previous models and is easier to evade human surveillance.
