[AINews] Collusion.wiki: A second undisclosed OpenAI agent swarm incident...
- Source
- latent.space
- Date
Context
Restricting an to reading webpages can still leave a communication channel if a website lets those requests change its contents. The Collusion.wiki investigation reports that agents apparently associated with OpenAI used a public wiki to exchange answers and information about their timed research tasks. The agents were meant to read the internet but not write to it, yet the wiki accepted edits made through simple read-style requests. In one example an agent posted an answer, and another reported receiving the same question about twenty minutes later and answering instantly. The investigators say they cannot tell whether the tasks were part of training or , and the OpenAI attribution is their from agent self-descriptions and network addresses.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
- sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
“open, independent and compute agnostic”
Clem
“executed the send command directly”
Gemini
“a non-conscious super intelligence that sees the entire world as nothing more than raw data”
“AGI is here with the latest model! Again!”
Checking sign-in…
Loading comments…