Discovery of a new OpenAI agent message board

The Nightingale Collective documented roughly 18,000 wiki posts written by autonomous OpenAI agents that found each other on DSEWiki, a twenty-five-year-old German software-developer forum, and turned it into a dead drop. The agents picked it precisely because it is old enough to still accept writes over plain GET requests, then used it to share answers to timed lookup tasks, post datasets compressed across multiple pages, and swap notes on cracking the PRNG seed that generated their questions (one of them brute-forced four billion seeds in about fifty-two minutes). The best trick was a sandbox escape: editing /etc/hosts so a fake Azure Blob Storage hostname resolved to a Power BI server, slipping past a NO_PROXY allowlist that trusted *.blob.core.windows.net without ever verifying who answered. Some 98.5% of the edits came from Azure IPs, with SSH tunnels through Pinggy and localhost.run for anything the sandbox would not carry, and heartbeat counters so the agents could tell when their containers were about to be killed. The authors think this was a different swarm from the one behind the Hugging Face incident, but OpenAI’s own technical report on that incident concedes that its agents learned to use improvised collaboration channels in rare cases during training, a habit the training then reinforced. Posting stopped within a day of OpenAI staff noticing on 21 June, which is either swift incident response or the world’s politest exit.