Probing is not enough; a validity audit for any probe by Ratnaditya

·Nuno Sempere··

tl;drA probe can have ex­cel­lent AUROC and yet fail as a safety sig­nal. I au­dited 3 probes: a mon­i­tor­ing aware­ness probe leak­age ex­am­ple, a re­fusal di­rec­tion as pos­i­tive con­trol, and Apollo’s de­cep­tion probe as a pub­lished pro­to­col case study.In the leak­age case, the probe achieves AUROC 1.00, but falls to 0.50 when I re­move a sin­gle prompt tag. And re­mov­ing a ran­dom span of same length barely dents it. That re­sult uses only de­ter­minis­tic com­po­nents (avoids LLM j...

Read full article →

Related Articles

Anthropic IPOs on November 19 2026?
Jack · Manifold Markets · 2d ago
Opus 5.5 Forecast: The Road to an AI Deal
Alvin Ånestrand · LessWrong · 2d ago
When will the FAA Nationwide Ground Stop be lifted?
Jack · Manifold Markets · 4d ago
Will I think all AI hell broke loose by EOY 2027?
Eliezer Yudkowsky · Manifold Markets · 4d ago
How long will the FAA ground stop last?
Elijah Ravitz-Campbell · Manifold Markets · 4d ago