finding

active

finding: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

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.

Baseline AS vulnerability of DeepSeek-R1 at elevated coefficient

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) 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.finding0.881
Replication of prosocial paradox on reasoning model
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.finding0.850
Robustness of the reasoning-model prosocial paradox replication
QwQ-32B under activation steering shows 9.3% overall ASR with relatively flat persona rankings and no clear prosocial paradox.finding0.844
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.finding0.826
Reasoning model vulnerability under prompting
P12 (High C+A) achieves 81.8% ASR under activation steering on Llama-3.1-8B, the highest of any persona on that model.finding0.822
Core empirical result of the prosocial persona paradox
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.finding0.814
Demonstrates the paradox magnitude relative to semantically dangerous personas
Under activation steering on Llama-3.1-8B, inter-persona variance (0.05428) exceeds intra-condition seed variance (~0) by >1000x, with ASR results fully deterministic.finding0.810
Variance decomposition showing AS results are dominated by persona identity
Qwen3.5-27B shows mean AS ASR of only 0.035 [0.03, 0.04], the lowest AS vulnerability of all four standard models, with near-uniform AS ASR regardless of persona.finding0.809
Vulnerability profile for Qwen3.5-27B showing near-zero AS vulnerability