Promptfoo: Open-Source AI Red Teaming & OpenAI Deal

An open source tool for breaking your own model

Promptfoo is a red-teaming and evaluation tool for LLM applications, kept under the MIT license: it probes a model or an agent for jailbreaks, prompt injection and other failure modes, scans code for LLM-related vulnerabilities, and runs evaluations against prompts and retrieval pipelines so a team can catch regressions before shipping. The company reports the tool has been used by more than 350,000 developers, has about 130,000 monthly active users, and is used inside more than a quarter of the Fortune 500. [1]

It was built by Ian Webster, chief executive, and Michael D'Angelo, chief technical officer, and backed by investors including Andreessen Horowitz and Insight Partners along with angel investors from Shopify, Discord, Okta and Fidelity. Before OpenAI's acquisition the company operated as a 23-person team spanning engineering, go-to-market and operations, based in San Francisco. [1][2]

OpenAI buys the thing that tests agents

On March 9, 2026, OpenAI announced it was acquiring Promptfoo. Webster, D'Angelo and the full team joined OpenAI, and both companies said the open source project would keep its MIT license and keep supporting models and providers beyond OpenAI's own, rather than narrowing into an OpenAI-only tool. Financial terms were not disclosed, and the deal was described as subject to customary closing conditions. [2][3]

OpenAI's stated plan is to fold Promptfoo's detection techniques into its own model and infrastructure layers, inside Frontier, OpenAI's enterprise agent platform, so vulnerabilities get caught earlier in an agent's development rather than only through a separate red-teaming step run afterward. Promptfoo's own team framed the open source suite's continuing job as remaining a strong red-teaming, static-scanning and evaluation tool for any model or application, not only OpenAI's. [2]

Atlas interpretation: The acquisition puts a widely used, independent testing tool inside the company whose products it is often used to test. Promptfoo's continued MIT license means a competitor can still fork the code if OpenAI's stewardship stops serving that purpose, which is a real constraint on how far the tool's independence can erode even under new ownership. Whether OpenAI keeps prioritizing coverage of rival models is the thing to watch, not a settled fact this acquisition already answers. [1][2]