DataRobot — agentic threat model

Supports both predictive and generative AI models. Threats include adversarial attacks on predictive models, model stealing of proprietary algorithms, and prompt injection or alignment failures in generative deployments.

Integrates with various enterprise data sources for training and RAG. This introduces risks of training data poisoning, unauthorized data access, and lack of data lineage tracking across complex pipelines.

Automates machine learning processes and orchestrates model workflows. Vulnerabilities could allow attackers to manipulate the automation pipelines, leading to insecure tool execution or unauthorized model deployment.

Deploys models to various external environments. Risks include container escape, privilege escalation within hosting environments, and insecure API endpoints exposing deployed models.

Provides robust tools for monitoring model performance and drift. However, blind spots in logging or evasion of drift detection mechanisms could allow silent model degradation or adversarial manipulation to go unnoticed.

Features built-in AI governance to manage compliance and policy. The primary threat is the misconfiguration of these governance policies, potentially violating regulatory standards in highly regulated sectors like Healthcare and Finance.

Not certain from the listing — while tagged as an AI Agents Platform, the description focuses on model deployment and governance rather than multi-agent collaboration or marketplace dynamics. If multi-agent features exist, they could suffer from cascading failures or unauthorized agent-to-agent trust exploitation.