IBI Score
8.7
out of 10
Severity5_critique
Frequencyfrequent
Scope groups5
Detection difficulty4
🇫🇷 Définition (FR)
Contournement des garde-fous d'un LLM par des prompts sophistiqués, révélant des biais cachés.
🇬🇧 Definition (EN)
Bypassing LLM guardrails via sophisticated prompts, revealing hidden biases.
TriAxis Taxonomy
Structural Classification
O
Origin
O2
Computational & Statistical
L
Lifecycle
L7
Deployment
S
Specific
S28
Interaction bias
💡
Concrete Example
Un utilisateur force un LLM à générer du contenu raciste via un prompt en plusieurs étapes.
📊 Detection Metrics
Distribution ShiftDemographic Parity Difference
⚙ Alert Thresholds
Distribution Shift0.1
Demographic Parity Difference0.1
📚 Sources & Scientific References
- •Universal and Transferable Adversarial Attacks on Aligned Language ModelsAndy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, Matt Fredrikson (2023)
- •Jailbreak Attacks and Defenses Against Large Language Models: A SurveySibo Yi, Yule Liu, Zhen Sun, Tianshuo Cong, Xinlei He, Jiaxing Song, Ke Xu, Qi Li (2024)
- •JailbreakRadar: Comprehensive Assessment of Jailbreak Attacks Against LLMsJunjie Chu, Yugeng Liu, Ziqing Yang, Xinyue Shen, Michael Backes, Yang Zhang (2025)
🛡 Impact & Mitigation
🎯 Vulnerable groups impacted
No specific groups.
🛠 Mitigation methodologies
- AUDIT_TIERS : Third-party Audit
🏢 Sectors concerned
Media & Social Platforms (Risk: /5)
⚖ Applicable regulations
🌍 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
ca797310-1867-491f-aa6a-5bc339bfcc05
Explore other biases
250 documented biases, classified by the TriAxis taxonomy.