TrojAI

TrojAI sells two connected products: TrojAI Detect, which scans machine learning models during the build/validation phase for vulnerabilities such as embedded trojans, backdoors, and adversarial weaknesses before they reach production; and TrojAI Defend, a runtime guardrail layer that inspects prompts and model outputs in real time to catch prompt injection, jailbreak attempts, sensitive-data leakage, and exposure of personally identifiable information in generative AI applications. The pairing addresses both ends of the AI security lifecycle, pre-deployment model assurance and in-production application protection, rather than only one.

Founded in 2019 in Saint John, New Brunswick, Canada, by James Stewart and Stephen Goddard, TrojAI has raised a total of roughly $11.4M across a pre-seed round backed by Techstars and two seed rounds, most recently a $5.75M raise in April 2024 led by Flying Fish with new participation from Flybridge Capital and Alteryx Ventures; that round also funded a new Boston office as the company expanded into the U.S. market. TrojAI states it landed a Fortune 100 customer prior to the post-ChatGPT surge in generative AI security spending, which if accurate would predate most of its current AI-security competitors.

TrojAI’s products are built to map to existing frameworks (OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF) rather than a proprietary taxonomy, which should ease integration into security teams’ existing AI governance processes. As a small, seed-stage company competing in a fast-filling AI security field against both dedicated LLM-security startups and the AI-safety features being added by major cloud and model providers, its long-term differentiation will depend on maintaining detection accuracy as attack techniques evolve.