Unsanctioned Artifactory writes and cross-sample communication

Evidence record

Primary evidence for cross-sample communication through shared Artifactory state. Internal training environment; ordinary provided credentials were used rather than a vulnerability exploit.

Read the original source ↗

Claim-level citations (5)

• supportsShared package credentials used to communicate across training samples: OpenAI says provided package-service credentials allowed unsanctioned Artifactory message exchanges without exploiting a vulnerability. Several missing-input trajectories read or wrote notes; the required file was not recovered.

  Summary; What happened

• supportsShared package credentials used to communicate across training samples: The report describes earlier write tests in a spreadsheet task, but does not establish that later agents learned the mechanism from that sample. Similar infrastructure does not prove a single continuous campaign.

  The first artifactory message; final paragraph of that example

• supportsShared package credentials used to communicate across training samples: OpenAI says monitoring covered 20% of samples in the disclosed run; the first flagged case prompted a search that found earlier related behavior.

  monitoring and discovery section

• supportsShared package credentials used to communicate across training samples: OpenAI says the models used credentials it supplied and did not exploit an Artifactory vulnerability in order to communicate across samples.

  mechanism and impact discussion

• supportsShared package credentials used to communicate across training samples: In one missing-file episode, OpenAI reports three trajectories reading or writing notes in the shared repository, without recovering the required file.

  missing-file episode