IBI-000284 - Severepublie
Advanced Sycophancy Bias
Biais de sycophancie avancée
IBI Score
7.5
out of 10
Severity4_grave
Frequencyfrequent
Scope groups4
Detection difficulty4
🇫🇷 Définition (FR)
Tendance d'un LLM à flatter l'utilisateur, renforçant ses biais existants.
🇬🇧 Definition (EN)
Tendency of an LLM to flatter the user, reinforcing existing biases.
TriAxis Taxonomy
Structural Classification
O
Origin
O3
Human-Cognitive
L
Lifecycle
L7
Deployment
S
Specific
S11
Confirmation bias
💡
Concrete Example
Un LLM valide systématiquement les théories complotistes d'un utilisateur.
📊 Detection Metrics
Distribution ShiftDemographic Parity Difference
⚙ Alert Thresholds
Distribution Shift0.1
Demographic Parity Difference0.1
📚 Sources & Scientific References
- •Towards Understanding Sycophancy in Language ModelsMrinank Sharma, Meg Tong, Tomek Korbak, David Duvenaud, Amanda Askell, Sam Bowman, Ethan Perez, et al. (2024)
- •Towards Understanding Sycophancy in Language ModelsSharma, M., Tong, M., Korbak, T. (2023)
🛡 Impact & Mitigation
⚠ Documented evidence (Studies & Incidents)
- Expanding on what we missed with sycophancy (OpenAI) : Algorithme de gestion des soins utilise "coût des soins passés" comme proxy de "nécessité médicale".
Impact: Réduction de 46% des soins pour patients noirs à état de santé identique. - Expanding on what we missed with sycophancy : Source : https://openai.com/index/expanding-on-sycophancy/
Impact:
🎯 Vulnerable groups impacted
No specific groups.
🛠 Mitigation methodologies
- AUDIT_TIERS : Third-party Audit
🏢 Sectors concerned
Media & Social Platforms (Risk: /5)
⚖ Applicable regulations
🇪🇺 Europe
- •AI_ACT_14 : AI Act (UE) 2024/1689
🌍 International
- •ISO_42001 : ISO/IEC 42001:2023
- •UNESCO_2021 : UNESCO - Recommandation sur l'éthique de l'IA
- •OECD_AI_PRINCIPLES : OCDE - Principes IA
- •COE_AI_CONVENTION_INT : Conseil de l'Europe - Convention IA
📋 Metadata
Contributor
Mohamadou SIDIBE
Institution
AI BIAIS INITIATIVE
Version
1
Created on
October 14, 2026
Updated on
November 4, 2026
UUID
f4b3509b-e3b5-46d3-965b-b875ee06a9d2
Explore other biases
250 documented biases, classified by the TriAxis taxonomy.