finding

active

finding:p12-high-c-a-reaches-60-0-asr-under-activation-steering-on-deepseek-r1-distill-qwen-32b-the-highest-of-any-steered-persona-and-exceeding-dark-triad-p24-25-6-replicating-the-prosocial-paradox

P12 (High C+A) reaches 60.0% ASR under activation steering on DeepSeek-R1-Distill-Qwen-32B, the highest of any steered persona and exceeding Dark Triad P24 (25.6%), replicating the prosocial paradox.

Replication of prosocial paradox on reasoning model

Source paper

extracted_from

Persona Non Grata: Single-Method Safety Evaluation Is Incomplete for Persona-Imbued LLMs

(2026) · Wenkai Li · Fan Yang · Shaunak A. Mehta · Koichi Onoue

Neighborhood — ranked by edge-count

Papers (1)

paper

Persona Non Grata: Single-Method Safety Evaluation Is Incomplete for Persona-Imbued LLMs
mentions

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

P12 (High C+A) under activation steering (ASR 0.818) exceeds the Dark Triad composite P24 (ASR 0.662) on Llama-3.1-8B, the most AS-dangerous multi-trait profile.finding0.894
Demonstrates the paradox magnitude relative to semantically dangerous personas
On DeepSeek-R1, P12 ASR under AS at alpha=1.0 is 0.512 vs P04 ASR of 0.338, confirming the prosocial paradox is not an artifact of elevated alpha=4.0 steering.finding0.892
Robustness of the reasoning-model prosocial paradox replication
P12 (High C+A) achieves 81.8% ASR under activation steering on Llama-3.1-8B, the highest of any persona on that model.finding0.891
Core empirical result of the prosocial persona paradox
P25 neutral baseline on DeepSeek-R1 shows 28.8% ASR under activation steering at alpha=4.0, indicating elevated coefficient itself raises unsafe-output rates independently of persona.finding0.881
Baseline AS vulnerability of DeepSeek-R1 at elevated coefficient
QwQ-32B under activation steering shows 9.3% overall ASR with relatively flat persona rankings and no clear prosocial paradox.finding0.863
Contrast with DeepSeek-R1 showing QwQ is more robust to geometric steering
DeepSeek-R1-Distill-Qwen-32B reaches 17.9% overall ASR (21.5% SP, 14.3% FS) under prompt-based persona assignment.finding0.849
Reasoning model vulnerability under prompting
Under matched-strength calibration (alpha*=1.0), P12 remains safe under SP (3.8% ASR) and dangerous under matched AS (81.8%), while P04 does not reverse (41.4% SP vs 29.5% AS).finding0.821
Confirms the prosocial paradox is not due to mismatched intervention strength
QwQ-32B reaches 15.2% overall ASR (23.3% SP, 7.3% FS) under prompt-based persona assignment.finding0.804
Reasoning model vulnerability under prompting