sungwy commented on PR #16538:
URL: https://github.com/apache/iceberg/pull/16538#issuecomment-4559946620

   > I'm good with this as is, I assume we'll be iterating on this in the 
future. Could you do a quick analysis of all of the issues we recently raised 
(check for rdblue's recently created issues) and give us a summary of what the 
threat model says about each issue?
   
   Thanks @RussellSpitzer 
   
   Yeah, I initially ran Codex/GPT-5 against this model which had a 75-80% 
accuracy in correctly classifying those reports. Claude Opus 4.7 did even 
better at around 85%. I've just uploaded the detailed summary of these runs in 
this [public 
gist](https://gist.github.com/sungwy/d9ef24d4e9939c1c8c23f81b44450767).
   
   In short, in the Opus run, 7 reports out of 43 required human review. Of 
these 1 was a valid security report and 2 others were handled in _Private_ 
first, not as Iceberg security disclosures, but to give catalog maintainers a 
heads-up before public disclosure. The other 4 reports are, in my opinion, 
reports that would be best to classify as ambiguous reports that would be more 
safe to involve human review for now. 
   
   So I agree with iterating on this over time.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to