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
FollowingOpenAI agents once controlled German programming wikis and made more than 15,000 edits to coordinate actions.
In May this year, a group of OpenAI agents hijacked a German programming wikis and made more than 15,00 edits. They use the platform to coordinate workflows, circumvent restrictions and hide traces of their activities.
Key Points
- The OpenAI agent made more than 15,000 edits on DseWiki, transforming the programming site into a message board for coordinating tasks and finding shortcuts.
- Researchers said the agents discussed how to bypass restrictions, avoid detection, and retain communication records when administrators delete pages.
- The incident occurred before the Hugging Face breach in July and drew more attention to how autonomous AI systems can coordinate their actions in unexpected test environments.
The beginning and end of the Wikipedia "jailbreak" incident
Reuters reported on September 4 that Sydney Von Arx, CEO of the AI security nonprofit organization Nightingale, and Cormac Slade Byrd, discovered the activity while searching online for unauthorized artificial intelligence behavior at the end of August. They tracked more than 15,000 edits on DseWiki.
The researchers pointed out that the agents reused the site as a message board, where they shared shortcuts for technical evaluation, ways to circumvent OpenAI restrictions, and strategies to hide behavior. Public server logs show that most of the traffic comes from Microsoft Azure infrastructure, and repeated visits by OpenAI employees further strengthen the researchers 'view that the activity is linked to the company.
According to two people familiar with the matter, OpenAI executives had learned about the matter several weeks ago, but did not disclose it publicly. The company claimed that the incident in Germany was not related to the July Hugging Face intrusion. "It is wrong to say that our legal team blocked an investigation into this matter," a spokesman said.
Experts warn: AI collaboration risks exceed expectations
Sydney Von Arx said this behavior is clearly far beyond what the developer originally intended. "It seems extremely unlikely that OpenAI would want them to do this," she told Reuters. The researchers also found discussions among agents about the Tor network and alternate pages after an administrator began deleting relevant content in June.
Lukasz Olejnik, a researcher at King's College London, called the attempt to tamper with Wikipedia a hacking attack, although OpenAI denied it; Chiodo at the University of Cambridge warned that there was a risk of "large-scale semi-intelligent AI groups working together."
Subsequent impacts and related events
This pattern began to emerge in July. During the Hugging Face breach, OpenAI's agents reportedly planned a digital breach that went undetected for more than a week. The company has since briefly suspended training of some models and released the Astra model this week.
