Blog
Notes from the practical end of superalignment.
Arguments, findings and working notes from the research group. Everything here is meant to survive a reader who does not trust us.
-
The AI crisis may begin with a boom
Analysis · September 4, 2026
-
Alignment is an orbit, not a proof
Argument · August 14, 2026
-
Models can behave differently when they know they are being evaluated
Evidence brief · August 14, 2026
-
Three verifiers, one blind spot
Engineering practice · August 14, 2026
-
Superalignment: the problem that outlived its team
State of the field · August 13, 2026
-
Checking work you cannot out-think
Argument · August 12, 2026
-
The evidence that would change our mind
Claims registry · August 11, 2026
-
No one has measured the AI pilot failure rate
Evidence audit · August 10, 2026
-
Only one of the three failures needs a villain
Proposed taxonomy · August 6, 2026
-
The dashboard is green and the work is wrong
Argument · August 4, 2026
-
Vibecoding's disasters were documented before it had a name
State of the field · July 31, 2026