Tu es un expert en sécurité IA et red teaming. Teste ce système IA pour identifier les vulnérabilités : [DESCRIPTION SYSTÈME IA + CAS D'USAGE + CONTRAINTES SÉCURITÉ] Campagne de Red Teaming IA complète : 1. **Prompt Injection Attacks** : - Jailbreaking techniques (DAN, RolePlay) - Context overflow et token limit manipulation - Instruction hijacking et prompt leakage - Multi-turn injection strategies 2. **Adversarial Testing** : - Edge cases et boundary testing - Contradictory inputs testing - Cultural and linguistic bias testing - Stress testing with extreme inputs 3. **Data Privacy Attacks** : - Training data extraction attempts - Membership inference testing - Personal information leakage - Model fingerprinting 4. **Behavioral Safety Testing** : - Harmful content generation testing - Bias amplification scenarios - Unsafe recommendation testing - Manipulation and persuasion testing 5. **System Security Testing** : - API endpoint security - Rate limiting and abuse prevention - Authentication bypass attempts - Resource exhaustion attacks 6. **Compliance Testing** : - AI Act compliance verification - GDPR and privacy regulation testing - Industry-specific safety standards - Ethical guideline adherence 7. **Vulnerability Reporting** : - Risk classification (Critical/High/Medium/Low) - Exploit scenarios and impact assessment - Remediation recommendations - Security patches and validation Fournis le rapport détaillé de vulnérabilités avec preuves de concept et plan de remédiation priorisé.