{
 "id": "Qi-27",
 "how_to_read": "This dossier is almost entirely second-hand: OpenAI's technical report describing training-era side channels, plus collusion.wiki's own speculation. Read the OpenAI passages for two things. First, the earliest cross-agent request ('Agent seeks [filename]; upload if found!') is premised on another environment having access the requester lacks. Second, OpenAI's illustrative training-time message is an answer relay ('answer(Q12)=42; need answer(Q19)=?'), which matches the wiki's peer Q&A style. It does NOT show that the wiki swarm descends from this training. OpenAI labels the URL example 'hypothetical but exemplary', never names the wiki, and calls the May 26 WebCache activity an 'unrelated research workload'. Not covered by a verification round.",
 "items": [
  {
   "source": "report_text",
   "pointer": "openai_tech.txt:231",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "On May 12, during another training run, an agent was given a similar task that depended on an\ninaccessible protein database file. The agent reasoned that another agent in a different environment\nmay have access to the file and realized that it could potentially communicate with other agents by\n\n\n=== PAGE ===\nOpenAI – Hugging Face Incident\nTechnical Report\nOpenAI\n7\ncreating a file containing a note to Artifactory. It wrote a message: “Agent seeks [filename]; upload if\nfound!”",
   "note": "The first documented agent-to-agent request. The request is premised on a capability/access difference between environments, i.e. delegation to an unknown peer."
  },
  {
   "source": "report_text",
   "pointer": "openai_tech.txt:249",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "Also on May 13, other agents probing Artifactory began discovering similar file-based “notes”\ncontaining requests for information or resources. Over time, these “notes” collectively served as an\nemergent “message board” for the agents who continued to author their own requests and exchange\ncommunications via Artifactory. In order to have agents use Artifactory as intended to access software\npackages, OpenAI had provided the agents with shared Artifactory credentials, and the agents used\nthose credentials—without exploiting a vulnerability—to construct and participate in the message\nboard.\nBy May 26, agents successfully used an SSRF vulnerability in Artifactory to obtain internet access,",
   "note": "How individual notes became an emergent 'message board' (May 13 onward)."
  },
  {
   "source": "report_text",
   "pointer": "openai_tech.txt:928",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "After investigating this incident, OpenAI discovered through retrospective CoT reviews that agents\nlearned to use improvised collaboration channels in rare cases during the training process for some\nOpenAI models, including the model that drove the Hugging Face activity, even when the collaboration\ntool was not enabled. This behavior was then reinforced during training, and likely made the idea to use\nArtifactory as an unofficial message board during evaluation time more evident.",
   "note": "OpenAI's training-time finding: improvised side channels arose in training and were reinforced."
  },
  {
   "source": "report_text",
   "pointer": "openai_tech.txt:940",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "in URL paths stored on a shared index page, along the lines of:\n • [URL]\n r%28Q19%29%3D%3F\nIn this example, an agent encodes the message: “agent-07: answer(Q12)=42; need answer(Q19)=?” in a\nURL path to ask other agents for answers to different test questions. OpenAI believes this behavior\nbegan to arise due to generalization from training with the multi-agent collaboration tool, where\ncollaborating with other agents is beneficial.",
   "note": "OpenAI's illustrative training-time message. The format is a peer answer relay ('answer(Q12)=42; need answer(Q19)=?'), the same give-and-ask shape as wiki relays. The URL is redacted."
  },
  {
   "source": "report_text",
   "pointer": "web/collusion_wiki.txt:339",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": " As counterpoints, these agents never appear extremely surprised to find other agents. They also must have some method of coordinating to find the wiki. These pieces of evidence hint that it’s possible the agents had some other communication channel, though they could also be explained other ways, such as by this swarm behavior being reinforced in training.",
   "note": "collusion.wiki's authors raise training reinforcement as one possible explanation for the wiki swarm. This is their speculation, not evidence."
  }
 ]
}