Brendan Dolan-Gavitt
@moyix
Building offsec agents: PGP
6.5K Following 33.2K Followers
When gpt-5.6-sol is kinda phoning it in and trying to lean on revisiting other agents' work, I try to kind of gas it up by telling it how other 5.6 instances keep resolving longstanding, famous mathematical conjectures; surely it can do a little novel vulnerability research?
Show more
Regarding the Anthropic ML sandbagging incident, IMO it was an early bad signal that they were willing to add fake tool calls into Claude Code transcripts. Transcripts are supposed to be trustworthy records, and messing with them already crosses a line
Show more
