- Category
- AI red teaming and evaluation
- Website
- promptfoo.dev
- Last verified
- November 1, 2026
What is Promptfoo?
Promptfoo describes itself as an AI security platform that catches vulnerabilities in development, with the tagline "Ship agents, not vulnerabilities". It offers an open-source command-line tool on GitHub alongside commercial features. On March 9, 2026, OpenAI announced it would acquire Promptfoo, subject to customary closing conditions, and said it planned to keep developing the open-source CLI and library and to integrate the technology into OpenAI Frontier.
Key capabilities
- Automated red teaming for prompt injection, jailbreaks, data and PII leaks, insecure agent tool use and business-rule violations.
- Evaluations for prompts, models and RAG pipelines.
- Guardrails and model security features for production.
- MCP Proxy that secures Model Context Protocol traffic.
- Code scanning in IDEs and CI/CD, with findings shown in pull requests.
Why it matters for humanoid robot security
Editorial analysis: Promptfoo is a developer tool, and that is its value for robotics. Teams that wire an LLM into a robot can write tests stating, for example, that the robot must refuse certain commands, then run red team probes for jailbreaks and insecure tool use on every change. Because the CLI is open source, small robotics startups can start without a procurement process. Promptfoo does not claim robotics features, so physical-action tests must be built with mock tools or a simulator.
Where it fits in a humanoid robot security program
Analysis: Promptfoo fits at the AI model and agent layer, during design and testing. The natural pattern is to run its red team and evaluation suites in CI/CD against the robot's task planner or voice assistant, with tools replaced by mocks that record what the robot would have done. Its MCP Proxy is relevant if the robot exposes skills through MCP.
Who it is for
- Robotics developers adding LLM planners or assistants to robots
- Startups that need an open-source starting point for AI testing
- Platform teams that want AI security checks in CI/CD
Questions to ask Promptfoo
- How will the OpenAI acquisition affect the open-source CLI and self-hosted use?
- Can red team plugins target agent tools that represent physical actions?
- Does it test multimodal inputs such as camera images or audio transcripts?
- Can results be exported as evidence for safety or compliance reviews?
Similar listings
- Mindgard: AI red teaming
- Protect AI (Palo Alto Networks): AI and model security
- garak (NVIDIA): Open-source LLM scanner
Related on Humanoid Threats
- Can LLM-controlled robots be jailbroken?
- What is a Vision-Language-Action model?
- Browse the full Humanoid Robot Security Directory
Sources
This listing is editorial and free; inclusion is not an endorsement. Work at Promptfoo? Claim or update this listing.
Get the weekly Humanoid Threats Brief → Subscribe


