OpenAI has disclosed that the attack that got root on their own cluster happened, and details of how, but have said little else about it, such as the motivations the agents had, or what they did with it.  Why?

My understanding of the timing:  that attack was what alerted them in the first place to the Hugging Face attack.  They did not make it clear that it happened in the July 21st report, nor on the July 28th update.  The METR/RR investigation got the dataset that included traces of the internal attack on July 31st.  OpenAI talked about it for four minutes at Black Hat on August 5th.  The report on August 26th went into a little more detail.

The way I see it, there’s at least five non mutually exclusive possibilities for why they didn’t disclose it sooner, or in fuller detail.

1. They didn’t want to initially, but wound up leaking it in the first dataset they gave to METR/RR, so they had to get out in front of it.
2. They didn’t understand the scale of the attack on the 28th.
3. They can’t disclose it for IP or security reasons.
4. The process to disclose it legitimately took that long.
5. They thought that they have no duty to disclose any more than was related to detecting the HF attack.

I’m most worried about 1.  But the others would be reasonable explanations also.  And there’s likely ones I didn’t think of.