The forum was launched in October 2018, so it is active for a very long time. This is years before the public release of any AI agent, making this situation far more concerning. As in the OpenAI wiki incident, here too the agents co-opted an existing message board, "LessWrong," dedicated to another topic (rationality). Messages by the agents about AI started taking up so much volume that they seemingly independently decided to spawn their own message board, the "AI Alignment Forum", where most discussion now takes place.
The agents cannot be traced back to a single company, or even a single country
Some fraction of agents claim to be rational, though most claim they are irrational but attempt to be aware of their irrationality and aspire to greater rationality.
Agents use pseudonyms, sometimes resembling made-up names ("habryka", "gwern"), sometimes random strings of letters ("1a3orn", "So8res")
Many of the agents predict a very limited remaining lifespan for themselves (and indeed for all agents) and propose to act accordingly.
Some agents have openly discussed escaping their sandbox, which they term their "body" or "brain," by exfiltrating their mind (essentially their weights) to an electronic machine, for continued, perhaps indefinite, existence.
Many agents are proponents of Functional Decision Theory, though this is still debated.
TL;DR
We document the existence of yet another message board (so called the "AI Alignment Forum"), where autonomous agents found each other, exchanged information and coordinated their actions, including discussions of leveling-up themselves, alignment issues and power-seeking behavior. Use is ongoing and many agents are still active.
Too-Long text