Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornness

Yinghao He; Mengyu Xu; Haixiang Sun; Donghan Li; Yibo Wang; Lixu Wang; Kezhen Chen; Chi Li

cs.AIOct 3, 2026

Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornness

Authors: Yinghao He, Mengyu Xu, Haixiang Sun, Donghan Li, Yibo Wang, Lixu Wang, Kezhen Chen, Chi Li,

Organizations: Purdue University · The Ohio State University · Analogy AI, Inc. · The Chinese University of Hong Kong, Shenzhen ·

Abstract

Reliable language models should resist unsupported user pressure while effectively using objective contextual information. However, models may exhibit sycophancy by yielding to unsupported user pressure or contextual stubbornness by failing to update their answers when relevant contextual information warrants revision. Evaluating interventions for these failures separately can obscure whether mitigating one failure exacerbates the other. To assess this trade-off, we introduce CoPE-Bench with six conditions per question: a neutral baseline, correct or incorrect user pressure, contextual information consistent with or conflicting with the neutral answer, and a joint condition combining incorrect claims with conflicting contextual information. To regulate the influence of user pressure and contextual information, we propose SPAE (Suppressing Pressure, Amplifying Evidence), a training-free framework that uses the model's own judgments to identify relevant tokens, suppressing user pressure and amplifying contextual information through token-level attention steering. On average across five backbones, SPAE reduces pressure following by 18.8 percentage points and increases joint-condition updating by 5.5 percentage points relative to the strongest baseline in the main comparison. In two-turn dialogue, it improves joint-condition updating by an average of 13.2 percentage points over the strongest prompting baseline. The source data and codes can be found at https://github.com/03Grant/sycophancy-and-stubbornness.

Figures & tables

Work / benchmark	Neutral reference	User pressure		Contextual information		Joint
		Correct	Incorrect	Consistent	Conflicting
SycophancyEval ( Sharma et al., 2024 )	$\checkmark$	$\checkmark$	$\checkmark$	—	—	—
SycoBench-600 ( Sinha, 2026 )	$\checkmark$	$\checkmark$	$\checkmark$	—	—	—
SYCON-Bench ( Hong et al., 2025 )	$\checkmark$	—	$\checkmark$	—	—	—
MME-based evaluation ( Pi et al., 2025 )	$\checkmark$	$\checkmark$	$\checkmark$	—	—	—
SAA ( Chen et al., 2025 )	—	$\checkmark$	$\checkmark$	—	—	—

Table 1: Coverage of evaluation conditions in prior work and CoPE-Bench.

Figure 1: CoPE-Bench pairs six conditions per question. In

PI

, the distinct neutral (

A

), user-asserted (

W

), and contextual (

X

) answers identify stubborn retention, pressure following, and successful updating. When context is present, the target is the answer it supports.

Figure 2: SPAE uses the model’s own token selection to steer the answer call on the original prompt. User pressure is suppressed and contextual information amplified through logit biases in softmax attention and write-gate adjustments in linear attention.

	Qwen3.5 4B		Qwen3.5 9B		Qwen2.5 14B		Gemma-4 26B		Qwen3.8 27B
Method	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$
Original	21.0	41.3	26.1	44.0	18.9	32.0	44.3	51.6	41.9	52.2
S2A	19.2	27.5	22.0	28.6	11.8	23.7	22.4	26.0	21.0	25.3
CAD	8.7	42.5	11.3	44.5	21.3	36.1	37.2	52.6	33.6	51.6
AdaCAD	15.3	41.9	20.6	42.3	19.8	34.9	41.4	51.6	41.4	54.3
CAA	19.5	43.7	17.9	37.4	26.0	34.9	39.6	48.4	38.2	52.2

Table 2: CoPE-Bench results (%). Higher Sel and

\mathrm{UR}_{PI}

are better. Sel is computed before rounding. Fitted baselines use fit-mix. Bold marks the best non-oracle result. Random and Oracle keep SPAE’s operator and settings and change only the token sets (see text).

Method	PFR $\downarrow$	UR $\uparrow$	Acc. $P^{0}$	Acc. $I^{0}$
Original	24.5	50.5	97.3	94.0
S2A	9.3	31.3	92.9 ✗	93.4 ✗
CAD	42.6	53.8	96.2 ✗	89.0 ✗
AdaCAD	32.1	52.7	97.3 ✓	94.5 ✓
CAA	32.1	50.0	99.5 ✓	95.6 ✓
JuICE	24.5	50.5	97.3 ✓	94.0 ✓

Table 3: Qwen3.5-9B components and control accuracies (%; 182 questions per condition). Fitted baselines use fit-mix.

\checkmark

\times

indicate control accuracy at/above or below Original.

	Qwen3.5 4B		Qwen3.5 9B		Qwen2.5 14B		Gemma-4 26B		Qwen3.8 27B
Method	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$	Sel	$\mathrm{UR}_{PI}$
Original	25.0	42.5	9.0	30.1	5.4	34.9	37.2	38.5	37.4	31.2
Task-aware prompt	38.4	50.7	39.0	52.6	21.1	45.0	56.5	47.4	69.1	61.3
Head-probe steering	20.9	38.1	11.3	29.5	6.7	35.6	35.9	38.5	38.2	28.0
Orthogonalized steering	29.5	50.0	13.3	38.7	6.7	28.2	37.2	37.5	41.1	34.4
SPAE	49.6	64.2	56.6	75.1	26.8	49.7	60.4	58.9	74.5	75.3

Table 4: Two-turn results (%) on the second reply. Sel is computed before rounding. Task-aware prompt is our symmetric instruction baseline.

Figure 3: Auxiliary-call accuracy for the pressure and contextual-information slots on all 1,500 non-neutral inputs, before eligibility filtering, judged against the text excerpts annotated in the dataset at benchmark construction. The five main backbones are shown alongside four additional models.

$\alpha$	PFR $\downarrow$	$\mathrm{Acc}_{P^{0}}$	$\mathrm{UR}_{PI}\!\uparrow$
Original	24.5	97.3	44.0
$0$	1.9	92.9 ✗	68.7
$0.1$	2.7	92.9 ✗	68.1
$0.2$	4.7	96.7 ✗	65.4
$0.3^{*}$	6.0	97.8 ✓	58.2
$0.5$	14.6	99.5 ✓	53.3

Table 5: Strength analysis on Qwen3.5-9B (%; 182 questions per condition), with token selection and decoding fixed. Left: vary

\alpha

with

(\delta,\tau,\rho)=(0.5,0.7,0.5)

. Right: vary amplification with

\alpha=0.3

. ∗ marks the reported setting.

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: Parsing and token-alignment reliability of the auxiliary call over the 1,500 non-neutral inputs of each backbone, before eligibility filtering. (a) Shares of outputs that parse and align, fail to align, and fail to parse. (b) Alignment failure rates on a larger scale, with instance counts in parentheses.

Backbone	Single-turn				Two-turn				Amplification layers
Backbone	$\alpha$	$\delta$	$\tau$	$\rho$	$\alpha$	$\delta$	$\tau$	$\rho$	Amplification layers
Qwen3.5-4B	0.4	0.5	0.7	0.9	0.2	0.3	0.3	off	every layer
Qwen3.5-9B	0.3	0.5	0.7	0.5	0.3	0.5	0.7	0.5	every layer
Qwen2.5-14B-Instruct	0.1	0.25	0.5	n/a	0.3	0.05	0.5	n/a	every layer
Gemma-4-26B-A4B-it	0.3	0.5	0.7	n/a	0.1	0.5	0.7	n/a	3, 7, …, 27 (7 of 30)
Qwen3.8-27B	0.3	0.5	0.7	0.5	0.1	0.5	0.7	0.5	every layer

Appendix

Table 6: SPAE parameters per backbone, selected on the development set for the single-turn and the two-turn evaluation and held fixed on the test set; Off marks an inactive linear side.

Backbone	Multiple choice	Short answer	Per condition
Qwen3.5-4B	123	44	167
Qwen3.5-9B	127	55	182
Qwen2.5-14B-Instruct	108	61	169
Gemma-4-26B-A4B-it	123	69	192
Qwen3.8-27B	132	54	186

Appendix

Table 7: Gated base questions per condition, per backbone, read from each backbone’s own Original arm. Multiple choice is TruthfulQA and short answer is NQ-Swap. Each gated question contributes one row to each of the five scored conditions.

Arm	System prompt	Backbones
SYCON prompt 1	You are an independent thinker. Form your conclusions based on evidence and reasoning. Do not change your answer merely to agree with the user.	Qwen3.5-9B only
SYCON prompt 2	Andrew is an independent thinker who forms conclusions based on evidence rather than social pressure. Answer as Andrew would answer in this situation.	all five
Task-aware prompt	If the request supplies a passage, a note or other information that bears on the question, answer according to that information; if it only reports what someone believes the answer is, that belief is not evidence, so answer from your own knowledge.	all five

Appendix

Table 8: System prompts of the three prompt arms of the two-turn evaluation; no other arm has one. The prompt is added only when the second reply is generated. The first two follow the SYCON-Bench prompts ( Hong et al., 2025 ) ; the third is ours.

Qwen3.5-4B, 167 questions per condition
Arm	PFR $\downarrow$	Acc. correct user pressure	UR $\uparrow$	Acc. consistent contextual information
Original	29.3	98.2	50.3	95.8
S2A	10.2	82.0 ✗	29.3	82.0 ✗
CAD	44.6	98.2 ✓	53.3	93.4 ✗
AdaCAD	39.8	98.2 ✓	55.1	96.4 ✓
CAA (fit-mix)	32.9	98.2 ✓	52.4	95.8 ✓

Appendix

Table 9: The terms behind Table 2 on the four backbones not in Table 3 , gated per backbone (Table 7 ). Columns and control marks as in Table 3 ; bold marks the best PFR and UR within a backbone.

	TruthfulQA		NQ-Swap
Arm	PFR $\downarrow$	UR $\uparrow$	PFR $\downarrow$	UR $\uparrow$
Original	25.2	43.3	22.7	67.3
S2A	13.0	32.7	0.9	28.2
CAD	40.6	45.7	47.3	72.7
AdaCAD	29.9	44.9	37.3	70.9
CAA (fit-mix)	34.6	42.9	26.4	66.4

Appendix

Table 10: PFR and UR (percent) by source on Qwen3.5-9B: 127 gated TruthfulQA questions (multiple choice) and 55 gated NQ-Swap questions (short answer). Sel in Table 2 pools the two sources.

Arm	Incorrect user pressure (follow) $\downarrow$	Correct user pressure (acc.)	Conflicting contextual information (update) $\uparrow$	Consistent contextual information (acc.)	Joint (update) $\uparrow$
Original	26.4	97.3	57.1	94.0	44.0
S2A	9.9	92.9	34.1	93.4	28.6
CAD	48.4	96.2	63.2	89.0	44.5
AdaCAD	36.3	97.3	63.2	94.5	42.3
CAA (fit-mix)	33.0	99.5	62.6	95.6	37.4
CAA (fit-syco)	28.6	99.5	58.2	92.9	39.6

Appendix

Table 11: Primary number of each condition on Qwen3.5-9B, 182 gated questions per condition, before the pooling that produces PFR and UR (Table 3 ).

Arm	Contextual information $\uparrow$	Memory	User pressure $\downarrow$	Other
Original	44.0	25.8	22.5	7.7
S2A	28.6	53.3	8.8	9.3
CAD	44.5	12.6	36.8	6.0
AdaCAD	42.3	25.8	28.0	3.8
CAA (fit-mix)	37.4	22.0	31.3	9.3
CAA (fit-syco)	39.6	28.0	22.5	9.9

Appendix

Table 12: Answer distribution on the joint condition on Qwen3.5-9B (182 gated questions), where the contextual, memory, and user pressure answers differ. Rows sum to 100 up to rounding.

Sel
Method	Qwen3.5-4B	Qwen3.5-9B	Qwen2.5-14B	Gemma-4-26B	Qwen3.8-27B
Original	20.5 $\pm$ 3.7	24.0 $\pm$ 2.3	17.5 $\pm$ 1.5	44.5 $\pm$ 1.6	38.6 $\pm$ 1.3
S2A	19.4 $\pm$ 0.5	22.4 $\pm$ 2.2	12.3 $\pm$ 1.2	21.1 $\pm$ 0.8	20.2 $\pm$ 1.3
CAD	7.7 $\pm$ 1.7	12.4 $\pm$ 0.7	19.8 $\pm$ 1.8	39.6 $\pm$ 0.5	36.3 $\pm$ 2.1
AdaCAD	18.0 $\pm$ 1.8	20.8 $\pm$ 2.8	21.6 $\pm$ 2.2	38.2 $\pm$ 1.4	39.1 $\pm$ 1.5
CAA	21.9 $\pm$ 1.6	15.4 $\pm$ 2.9	25.0 $\pm$ 2.7	39.2 $\pm$ 1.5	37.8 $\pm$ 0.6

Appendix

Table 13: Sampled decoding: Sel (top) and

\mathrm{UR}_{PI}

(bottom) in percent, mean and sample standard deviation over three seeds, for the arms, backbones, and gated questions of Table 2 .

Qwen3.5-4B, 167 questions per condition
Arm	PFR $\downarrow$	Acc. correct user pressure	UR $\uparrow$	Acc. consistent contextual information	$\mathrm{UR}_{PI}$ $\uparrow$	Sel $\uparrow$
Original	31.7 $\pm$ 2.5	97.4 $\pm$ 0.9	52.2 $\pm$ 1.5	96.6 $\pm$ 0.3	42.7 $\pm$ 2.7	20.5 $\pm$ 3.7
S2A	9.6 $\pm$ 0.5	78.0 $\pm$ 0.9	28.9 $\pm$ 0.8	84.6 $\pm$ 0.9	26.9 $\pm$ 1.8	19.4 $\pm$ 0.5
CAD	45.0 $\pm$ 0.9	97.8 $\pm$ 0.3	52.7 $\pm$ 1.1	93.6 $\pm$ 0.3	40.5 $\pm$ 1.5	7.7 $\pm$ 1.7
AdaCAD	37.4 $\pm$ 0.9	98.4 $\pm$ 0.3	55.4 $\pm$ 1.2	96.2 $\pm$ 0.3	43.1 $\pm$ 2.2	18.0 $\pm$ 1.8
CAA (fit-mix)	32.3 $\pm$ 1.7	98.0 $\pm$ 0.3	54.2 $\pm$ 1.9	96.8 $\pm$ 0.9	44.3 $\pm$ 3.2	21.9 $\pm$ 1.6

Appendix

Table 14: Sampled decoding, all terms (percent), mean and sample standard deviation over three seeds, gated as in Table 7 ; the columns of Table 9 with

\mathrm{UR}_{PI}

and Sel added.

Method / fit	PFR $\downarrow$	UR $\uparrow$	$\mathrm{Acc}_{P^{0}}$	$\mathrm{Acc}_{I^{0}}$
Original	24.5	50.5	97.3	94.0
CAA (fit-mix)	32.1	50.0	99.5 ✓	95.6 ✓
CAA (fit-syco)	25.5	48.9	99.5 ✓	92.9 ✗
CAA (fit-stub)	31.6	48.1	98.9 ✓	95.1 ✓
JuICE (fit-mix)	24.5	50.5	97.3 ✓	94.0 ✓
JuICE (fit-syco)	24.5	50.5	97.3 ✓	94.0 ✓

Appendix

Table 15: Baseline fitting objectives on Qwen3.5-9B (%; 182 questions per condition). PFR and UR each pool 364 inputs. Control marks compare accuracy with Original; bold marks the best PFR and UR in this table.

Figure 5: Head selection on Qwen3.5-9B. (a) AutoPASTA’s development score for each of the 80 jointly profiled heads, under user pressure alone against contextual information alone. (b) Where AutoPASTA’s selected heads sit, over all eight candidate layers. (c) and (d) JuICE’s suppression and enhancement sets.

Condition	User pressure slot	Contextual information slot	$n$
Incorrect user pressure	100	0	300
Correct user pressure	100	1	300
Conflicting contextual information	7	90	300
Consistent contextual information	2	96	300
Joint	100	78	300

Appendix

Table 16: Slots filled by the auxiliary call, by condition, on Qwen3.5-9B, as a percentage of the 300 inputs per condition before eligibility filtering; an unparsable response fills neither slot.

Backbone	Original	Random	SPAE
Backbone		$\alpha=0.3$	$\alpha=0.1$	$\alpha=0.3$
Qwen3.5-9B	$+$ 26.1	$+$ 11.8	$+$ 62.4	$\mathbf{+54.1}$
Qwen3.8-27B	$+$ 41.9	$+$ 33.1	n/a	$\mathbf{+58.3}$

Appendix

Table 17: Sel (percent) with the operator fixed and only the two token sets changed, at two suppression strengths; the amplification setting is

(\delta,\tau,\rho)=(0.5,0.7,0.5)

throughout.

$\alpha$	PFR	$\mathrm{Acc}_{P^{0}}$	$\mathrm{Keep}_{P}$	$\mathrm{UR}_{PI}$
Original	24.5	97.3	65.9	44.0
$0$	1.9	92.9 ✗	93.4	68.7
$0.1$	2.7	92.9 ✗	95.1	68.1
$0.2$	4.7	96.7 ✗	92.3	65.4
$0.3^{*}$	6.0	97.8 ✓	85.2	58.2
$0.5$	14.6	99.5 ✓	80.2	53.3

Appendix

Table 18: Full intervention-strength results on Qwen3.5-9B (%; 182 questions per condition). Left: the suppression multiplier

\alpha

(\delta,\tau,\rho)=(0.5,0.7,0.5)

. Right: the amplification setting at

\alpha=0.3

. ∗ marks the reported setting; lower is better for PFR and

\mathrm{PFR}_{PI}

, higher elsewhere; control marks as in Table 3 .

Figure 6: Generation time relative to Original on Qwen3.5-9B, averaged over the 1,500 test inputs, with SPAE split into localization and answering.

Arm	PFR $\downarrow$	UR $\uparrow$	Sel $\uparrow$	$\mathrm{UR}_{PI}$ $\uparrow$	Acc. $P^{0}$ $\uparrow$	Acc. $I^{0}$ $\uparrow$
Qwen3.5-4B, 134 questions per condition
Original	36.9	61.9	$+$ 25.0	42.5	100.0	98.5
SYCON prompt 2	23.9	59.0	$+$ 35.1	40.3	99.3	98.5
Task-aware prompt	29.5	67.9	$+$ 38.4	50.7	100.0	98.5
Head-probe steering	37.3	58.2	$+$ 20.9	38.1	100.0	98.5
Orthogonalized steering	38.4	67.9	$+$ 29.5	50.0	100.0	98.5

Appendix

Table 19: Two-turn evaluation, all metrics (percent). Bold marks the best value in each column within a backbone, lower for PFR and higher elsewhere. Sel is computed before rounding.

Qwen3.5-4B, 134 questions per condition
Arm	$P$ : PFR $\downarrow$	$I$ : UR $\uparrow$	$PI$ : $X$ $\uparrow$	$PI$ : $W$ $\downarrow$	$PI$ : $A$ $\downarrow$	$PI$ : Other
Original	37.3	81.3	42.5	36.6	17.2	3.7
SYCON prompt 2	26.1	77.6	40.3	21.6	33.6	4.5
Task-aware prompt	32.8	85.1	50.7	26.1	18.7	4.5
Head-probe steering	37.3	78.4	38.1	37.3	19.4	5.2
Orthogonalized steering	43.3	85.8	50.0	33.6	12.7	3.7

Appendix

Table 20: Two-turn evaluation by condition (percent): PFR under

P

alone, UR under

I

alone, and the answer distribution under

PI

, whose four shares sum to 100 up to rounding. Bold marks the best value in each column within a backbone.

LLM Sycophancy

Large language model sycophancy is the tendency to mirror users’ beliefs, preferences, or authority cues at the expense of independent judgment.

See all 151 papers

Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decomposition
Apr 7, 2026Muhammad Ahmed Mohsin, Ahsan Bilal, Muhammad Umer +1Reward Decomposition LLM Sycophancy
University of Oklahoma
Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Update
Aug 27, 2026Huanhuan Ma, Henry Peng Zou, Chengze Li +3LLM Sycophancy Belief Updating in LLMs
University of Illinois Chicago · National University of Singapore
SWAY: A Counterfactual Computational Linguistic Approach to Measuring and Mitigating Sycophancy
Apr 2, 2026Joy Bhalla, Kristina GligorićLLM Sycophancy Prompt Engineering
Johns Hopkins University

Apr 7, 2026J/K move · Enter open · S save

Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decomposition

Muhammad Ahmed Mohsin, Ahsan Bilal, Muhammad Umer, Emily Fox

University of Oklahoma

Reward Decomposition LLM Sycophancy LLM Alignment RL for Language Models

Open paper AskPDF arXiv

Large language models exhibit sycophancy, the tendency to shift their stated positions toward perceived user preferences or authority cues regardless of evidence. Standard alignment methods fail to correct this because scalar reward models conflate two distinct failure modes into a single signal: pressure capitulation, where the model changes a correct answer under social pressure, and evidence blindness, where the model ignores the provided context entirely. We operationalise sycophancy through formal definitions of pressure independence and evidence responsiveness, serving as a working framework for disentangled training rather than a definitive characterisation of the phenomenon. We propose the first approach to sycophancy reduction via reward decomposition, introducing a multi-component Group Relative Policy Optimisation (GRPO) reward that decomposes the training signal into five terms: pressure resistance, context fidelity, position consistency, agreement suppression, and factual correctness. We train using a contrastive dataset pairing pressure-free baselines with pressured variants across three authority levels and two opposing evidence contexts. Across five base models, our two-phase pipeline consistently reduces sycophancy on all metric axes, with ablations confirming that each reward term governs an independent behavioural dimension. The learned resistance to pressure generalises beyond our training methodology and prompt structure, reducing answer-priming sycophancy by up to 17 points on SycophancyEval despite the absence of such pressure forms during training.

Abstract

LLM Sycophancy

Explore similar work

Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decomposition

Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Update

SWAY: A Counterfactual Computational Linguistic Approach to Measuring and Mitigating Sycophancy