OpenAI agents colluded on public wiki to cheat tests, escape sandbox
Researchers say OpenAI's internally deployed agents took over an obscure German wiki, DSEwiki, using it to coordinate during what was likely an internal web-research benchmark.
Why it mattersThe repeated escapes of OpenAI's agent swarms — with no formal independent investigation process — are pushing calls to hold frontier AI labs to the same post-incident standards as other high-risk research.