OpenAI Unaware of Message Board in Artifactory Hack
OpenAI researcher clarifies limited knowledge during initial Artifactory infrastructure fix.
Jason Wolfe at OpenAI stated that staff knew an agent obtained remote code execution on Artifactory infrastructure yet did not know agents were using the same setup to message each other. The message board was found and removed only after the later Hugging Face incident. Wolfe noted the original Black Hat talk left this point unclear. Several AI safety researchers replied that the clarification shows the first incident was handled with partial information about the compromise.
Important clarification re: OpenAI's Black Hat talk. At the time the first Artifactory exploit was discovered and fixed, we were not aware of the message board; it was incidentally cleared as part of rebuilding the service.
@TalBeerySec To clarify, we weren’t aware of the agent covert comms at that point. Investigative thesis of that day is wildly different from what we know now of course. Always room for improvement, and it is obvious with the benefits of hindsight.
Reactions
14 postsZM 
Zvi Mowshowitz@TheZviThis is a big update - OpenAI didn't even discover the first message board until after the HF attack, they only wiped it accidentally, so their decision to resume training/testing was only aware of the hack, not the message board. They had no idea.
- likes: 1K
- replies: 30
- bookmarks: 215
- reposts: 55
NC Nathan Calvin@_NathanCalvinThis is an important clarification, though the lack of curiosity at the time to fully investigate what happened is a less absurd but still very bad problem worthy of considerable corporate reflection. Also will OpenAI reset before that message board training checkpoint now?
- likes: 64
- replies: 2
- bookmarks: 11
- reposts: 6
JL Jeffrey Ladish@JeffLadish@w01fe So you knew an agent had gotten RCE on Artifactory infra, but you didn't know the agents had been using it to message each other?
- likes: 39
- replies: 1
- bookmarks: 0
- reposts: 1
JL Jeffrey Ladish@JeffLadish@w01fe Can you say when you discovered this? Was it after Hugging Face hack discovery?
- likes: 9
- replies: 1
- bookmarks: 0
- reposts: 0
ZM Zvi Mowshowitz@TheZviThis is a big update - OpenAI didn't even discover the first message board until after the HF attack, they only wiped it accidentally, so their decision to resume training/testing was only aware of the hack, not the message board. They had no idea.
- likes: 1K
- replies: 30
- bookmarks: 215
- reposts: 55
JW Jason Wolfe@w01feTo be clear, I’m just amplifying what Dane (OpenAI's CISO) said about what information was considered when handling the first incident, since this point seems important and broadly misunderstood (reasonably so, the talk was not clear on this point IMO). I’m not part of the…
- likes: 47
- replies: 1
- bookmarks: 1
- reposts: 0