TrojAI

TrojAI is an enterprise AI security platform for organizations deploying AI models, LLM applications, and AI agents that need protection against adversarial attacks and unsafe behavior.

Key capabilities include:

• Build-time automated red teaming (TrojAI Detect) to uncover vulnerabilities before deployment, including risky behaviors that can lead to data exposure or compromised model behavior.
• Runtime protection with an AI firewall (TrojAI Defend) that monitors inputs and outputs and helps detect, block, redact, alert, and log active threats.
• Controls for prompt injection and jailbreaking, tool misuse, and unsafe agent actions across complex enterprise workflows.
• Support for any model architecture, multi-cloud, and enterprise environments, including flexible deployment options such as self-hosting.

Pricing: Unknown.

Notable for covering both stages of the AI lifecycle (pre-deploy testing plus in-production enforcement) and for its connection to security research efforts such as NIST’s TrojAI Leaderboard work on detecting trojans in trained models.