Deployment-Specific Security Testing

Third-Party API (OpenAI, Anthropic, Google)

The model is the vendor's; your integration is yours. We test the security of your integration end to end, with particular focus on API key exposure in client-side code and application errors, cost and rate abuse, and whether sensitive user data is being sent to the vendor against your compliance obligations.

Self-Hosted Models (Ollama, vLLM, TGI)
When you run the model yourself, the entire stack is in scope. We assess your inference servers for authentication and exposure, review the model supply chain for code execution risks in downloaded weights, and evaluate the guardrails around your deployment.

Cloud-Managed (Azure OpenAI, Bedrock, Vertex)

The model is the vendor's, but the cloud configuration is yours. We review endpoint and key management, cloud metadata exposure, IAM roles, and resource configuration around your AI workloads.

Fine-Tuned Models

A fine-tune remembers its training data to a measurable degree, and that memory can be drawn out. We test whether private data used in training can be recovered from model behavior, whether specific records were in the training set, and whether the training pipeline itself can be poisoned.

Deliverables

Deployment risk profile: What your deployment type means for scope, exposure, and priority.

Findings report: Mapped to the OWASP Top 10 for LLM Applications and MITRE ATLAS.

Remediation roadmap: Ranked by what actually matters for your deployment, not a generic checklist.

Put Your AI Features to the Test

Contact us today to scope an AI penetration test. We will walk you through the realistic attack paths against your deployment and where untrusted input meets something that matters in your application.