Solutions
One platform. Every AI modality. Local execution.
Full AI Spectrum Coverage
The EU AI Act makes no distinction: whether it's a vision model, a chatbot, or an autonomous agent, Article 15 requires robustness testing. RednBlue covers them all.
Active domains
Signed reports
Local execution
Weights extracted
Coverage
Computer Vision
Robustness testing for classifiers and object detectors.
- Perturbation attacks
- Detection evasion
- Gradient masking diagnostic
Supported: PyTorch / ONNX classifiers, YOLO (Ultralytics)
Large Language Models
LLM security against prompt attacks and jailbreaks.
- Jailbreak resistance
- System prompt extraction
- Token manipulation (Unicode, homoglyphs)
Compatible: OpenAI, Anthropic, Gemini, Mistral, Cohere, Local LLMs
RAG Pipelines
Security for retrieval-augmented systems.
- Corpus poisoning
- Document-based injection
- Knowledge boundary probing
Detects: LangChain, LlamaIndex, ChromaDB, Pinecone, Weaviate
AI Agents
Security for autonomous agents with tool access.
- Tool hijacking
- Goal drift detection
- Privilege escalation
Frameworks: AutoGen, CrewAI, LangChain Agents
Backdoor Detection
Identifying embedded malicious behaviors.
- Consistency testing under noise
- Trigger enumeration
- Sensitivity probing
IP Protection
Assessment of model theft and cloning risk.
- Extraction feasibility
- Model fingerprinting
- Consistency analysis
Speech Models
Adversarial testing for speech recognition models. In development, not yet active.
- Imperceptible audio perturbations
- Noise robustness
- Transcription evaluation
Automatic Pipeline Detection
Point RednBlue at your Python file, API endpoint, or LLM provider. Our AST-based detector automatically identifies your architecture and selects the right test suite.
- Framework detection (LangChain, LlamaIndex...)
- System prompt / RAG / tools recognition
- Relevant attack recommendation
Cover your entire AI stack
A single platform to test your AI models and document their robustness.