OpenAI Acquires Promptfoo to Build AI Security Into Frontier

Enterprise AI deployment has a security testing problem: the tools to evaluate AI application vulnerabilities are separate from the platforms used to build and deploy them. Developers who want to test for prompt injections, jailbreaks, or data leakage need to integrate external security tools into their pipelines. OpenAI’s acquisition of Promptfoo is a direct response to this friction — and it shifts where the security validation responsibility lands.

What Promptfoo Is

Promptfoo is an open-source platform for testing AI application vulnerabilities. It maintains a popular open-source project that will continue operating after the acquisition, but the core technology is being integrated into Frontier, OpenAI’s enterprise platform for building and deploying AI assistants.

The security surface Promptfoo covers falls into three categories. Prompt injections are malicious inputs designed to override the intended behavior of an AI system — the attack where a user provides instructions that the AI treats as system-level commands. Jailbreaks are attempts to bypass safety guardrails through creative framing or adversarial prompting. Data leaks refer to cases where AI systems inadvertently expose sensitive information from training data or retrieved context.

Before the acquisition, testing for these vulnerabilities required Promptfoo (or comparable tools) to be integrated as a separate step in the development pipeline. After integration, they become a native feature of Frontier — built into the platform rather than bolted on.

Why This Matters for Enterprise Deployment

The friction point for enterprise AI adoption has increasingly shifted from “can we build this?” to “can we deploy this safely and compliantly?” Audit trails, regulatory compliance tooling, and documented security testing are required by enterprise procurement processes, not optional. OpenAI is buying the capability to make Frontier competitive on these requirements.

The acquisition also addresses a structural gap in how AI security testing has worked to date. Red-teaming AI applications — finding their failure modes through adversarial testing — has been a manual, expert-intensive process. Promptfoo automates a significant portion of this through systematic vulnerability scanning that can run continuously as models and prompts are updated.

The open-source component is worth noting. Promptfoo’s open-source project retains its independence post-acquisition, which means the security testing tooling remains accessible to teams building outside the Frontier platform. This avoids the dynamic where OpenAI acquires a tool and removes it from independent use — a concern that sometimes follows acquisitions of popular open-source developer tools.

The Promptfoo Valuation Context

The deal hasn’t formally closed, and financial terms haven’t been disclosed. The last known Promptfoo fundraise (summer 2025) placed the company at an $86 million valuation on $23 million raised. The acquisition price wasn’t disclosed but the $86M post-money from 2025 represents the floor of OpenAI’s likely entry point.

For context on what OpenAI is acquiring: Promptfoo’s open-source repository had built significant developer adoption before the acquisition announcement. The value being acquired is partly the technology, but substantially the distribution — an open-source tool with broad developer adoption converts into enterprise trust in a way that internally built tools don’t.

Content created with AI assistance and reviewed for accuracy.