OpenAI aims to expand its security and testing capabilities for AI applications through the acquisition of Promptfoo. The company announced it wants to integrate the startup’s technology into OpenAI Frontier, a platform that helps organizations build and manage AI agents for business processes.
The acquisition is part of a broader trend in which companies are increasingly using AI agents in operational workflows. This is creating a growing need for systematic evaluation, security, and control mechanisms. Organizations want to test how AI systems behave in advance. They also want to detect risks before systems go into production and maintain documentation for oversight and compliance.
Promptfoo, founded by Ian Webster and Michael D’Angelo, develops tools that enable developers and security teams to test AI applications for vulnerabilities. According to OpenAI, the platform is used by more than a quarter of Fortune 500 companies. The team also maintains an open-source command-line interface and library for evaluating and red-teaming applications that use large language models.
Promptfoo started as an open-source testing framework
The Register adds that Promptfoo started in 2024 as an open-source framework for evaluating prompts and model behavior. It later grew into a commercial platform that allows developers and enterprise security teams to test AI applications running on large language models or other generative AI systems.
The technology focuses on risks arising from the deployment of AI models in production environments. These include prompt injections, data leaks, jailbreak attacks, and unsafe execution of external tools. The platform enables systematic testing of how AI systems respond to structured inputs, adversarial prompts, and scenarios that simulate realistic use.
The testing framework allows teams to define prompts, expected outputs, and evaluation criteria in configuration files. These tests can then be automatically run against one or more AI models or applications. The results are compared to predefined rules or assessment functions to determine whether the system is behaving as expected.
Promptfoo also supports integrations with commonly used development tools and APIs. This allows evaluations to be performed locally and within CI/CD pipelines or automated deployment processes. The tooling includes features for prompt testing, red team simulations, and dashboards that provide insight into model behavior in different circumstances. It also supports regression testing, which allows organizations to compare outputs from different model versions or configurations before rolling out updates.
Integration into OpenAI Frontier
OpenAI wants to integrate this functionality directly into Frontier. This should enable organizations to test agent behavior more effectively, detect risks earlier in the development process, and document changes to AI systems more effectively for governance and compliance.
According to Srinivas Narayanan, CTO for B2B applications at OpenAI, Promptfoo specializes in evaluating and securing large-scale AI systems. By incorporating the technology into Frontier, the company expects that organizations will be able to develop and manage reliable AI applications more easily.
Webster states that Promptfoo arose from developers’ need for practical tools to make AI systems secure. Now that AI agents are increasingly linked to business data and external systems, he believes it is becoming more complex to monitor and validate their behavior. By joining OpenAI, the team wants to accelerate that development.
Prior to the acquisition, Promptfoo had raised $23.6 million in funding, including an $18.4 million investment round in July. Investors include Insight Partners and Andreessen Horowitz.
The transaction is still subject to completion and is contingent upon customary conditions for an acquisition.