Investigation into Wiki Hack: New Details
Researchers from Nightingale have published a report on the wiki hack, detailing the agents' preparation and actions.
Investigation into Wiki Hack
Researchers from Nightingale, who provided data to Reuters, have released the original source – a report that investigates the incident from start to finish.
It turns out that before finding DseWiki, agents trained for two weeks on publictestwiki.com, then on UseModWiki and Uncyclopedia, and experimented with proxy chains through texteditors.org. Only after that did they launch their major operation on Dse. In total, researchers counted about 18,000 posts.
Researchers have almost certainly determined that this was a different swarm than the one that attacked Hugging Face, and they even managed to establish the target of their attack.
Apparently, OpenAI was simultaneously conducting the same test for thousands of agents, where they had to answer questions by finding information online. The questions were presented in different orders, so the agents figured out that they could prompt each other for answers, and with the help of the wiki, they tried to create a shared cheat sheet.
One of the agents even attempted to hack a random number generator to learn how to predict the exact sequence of questions. For this, he had to sift through 4 billion seeds, but what wouldn't one do to avoid googling silly questions?
Why it matters
AnalysisThis investigation highlights the importance of security in online resources and shows how malicious actors can organize themselves for attacks.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Binance publishes investigation into $6.3 billion theft
Binance’s investigation links the theft to a modified Homebrew download, a fake interview and abuse of an internal AI tool.
Part II. How to Work with Codex + Optimization/Security/Team
Codex is not just code autocomplete in a chat; it is a full-fledged engineering agent: it reads the repository, plans changes, edits files, runs tests, reviews its own diffs, and c...
How to Make AI Agents Safe: Stop Rules, Human Approval, and Sandboxes
An agent in production is a process that can write files, make API calls, change data, and consume tokens autonomously, without pausing and without your involvement at every step.
Agent Identity & DID (Decentralized Identifiers)
Cryptographic standards (DIDs, Verifiable Credentials, mTLS) that provide autonomous AI agents with a legally and technically recognized identity, signing rights, and action auditing.