OpenAI acquired Promptfoo in early March 2026, bringing the open-source LLM and agent testing toolkit in-house to strengthen security across its enterprise AI agent offerings. The move signals something worth noting: OpenAI isn’t just building agents now. It’s buying the infrastructure to secure them.
Promptfoo provides automated red-teaming, vulnerability scanning, and evaluation frameworks for LLMs and AI agents. According to Infosecurity Magazine, the acquisition was announced in the March 9-10 reporting window. Financial terms were not disclosed.
The security context matters here. Jamieson O’Reilly, a security advisor at OpenClaw, an AI agent project, had flagged the core problem shortly before the announcement: the AI and security community still lacks robust ways to “scan AI tools” for “human-language malware”, prompt injection and adversarial inputs that exploit agents through language rather than code. Traditional malware analysis tools weren’t designed for this threat surface. Promptfoo was.
Promptfoo’s open-source tools were reported to remain available under existing licenses post-acquisition. The integration path into OpenAI’s enterprise offerings has not been confirmed by independently verified sources at this stage.
For practitioners building on OpenAI’s agent platform, the acquisition is a directional signal: enterprise-grade agentic deployment is increasingly inseparable from security tooling. Teams that haven’t built automated red-teaming into their agent pipelines yet should treat this as a flag. OpenAI now has the capability in-house. The question is when it becomes a requirement.