Security and compliance

Threat modelling

A threat model has been conducted using the Open Worldwide Application Security Project (OWASP) Top 10 for LLMs and GenAI, identifying potential vulnerabilities that are specific to machine learning systems and AI-powered analytics platforms.

The following table summarizes the key threat categories identified during the threat modeling process, along with example attack scenarios and the mitigations implemented to address each risk:

Threat category	Example attack	Mitigation implemented
Prompt Injection	Malicious input alters model output	Input sanitization, Bedrock guardrails
Data Poisoning	Corrupted training data skews predictions	Data quality monitoring, validation jobs
Insecure Output Handling	Harmful model output stored or executed	Output sanitization, guardrails
Model Tampering	Unauthorized changes to deployed models	Model lifecycle monitoring, IAM restrictions
Sensitive Data Exposure	PII leaked in model responses or logs	Logging review, PII masking, tenant isolation
Denial of Service	High-volume requests overload the system	Rate limiting, throttling, and caching
Supply Chain Compromise	Vulnerable ML images or repos exploited	SAST, DAST, and CVE scans on all artifacts

Regulatory compliance frameworks

EU AI Act classification

Classified as a low-risk AI system under the EU AI Act due to analytical-only functionality with no autonomous decision-making capabilities. Human oversight is required for all outputs and recommendations.