OpenAI to buy Promptfoo for enterprise agent security

Giving AI agents access to company databases and internal tools is thrilling right now, but it is also an absolute nightmare for security teams. A single sneaky prompt injection could trick an autonomous agent into leaking sensitive records or misusing internal tools.

To convince big businesses that AI coworkers are actually safe to hire, OpenAI is buying one of the top security toolkits in the space.

OpenAI announced an agreement to acquire Promptfoo, an AI security platform built to detect and fix vulnerabilities in AI systems. Once the deal closes, Promptfoo’s technology will be embedded directly into OpenAI Frontier, the company's platform for building and operating enterprise AI coworkers.

Automated red-teaming out of the box

Promptfoo was founded by Ian Webster and Michael D'Angelo. Despite keeping a relatively low public profile, its software has already been adopted by more than 25 percent of Fortune 500 companies.

Developers widely know Promptfoo for its popular open-source command line tool and evaluation library, which simulate adversarial attacks and benchmark large language model applications.

OpenAI confirmed that it will keep maintaining the open-source project while weaving the core enterprise tech into Frontier. The goal is to make automated security testing and red-teaming a native part of the development cycle.

That means teams building on Frontier will get built-in protections against prompt injections, jailbreaks, data leakage, and unauthorized tool calls, alongside compliance reporting to keep internal auditors happy.

Why it matters

We think this is a telltale sign of where the enterprise AI race is really being fought today. Raw model intelligence is no longer the main bottleneck keeping agents out of production. The real blocker is enterprise compliance and risk teams refusing to sign off.

By pulling Promptfoo into Frontier, OpenAI is making safety a turnkey feature rather than an afterthought. Developers won't need to cobble together third-party security pipelines just to run basic adversarial checks on their agents before deployment.

Strategically, it is a smart grab that tightens OpenAI's grip on enterprise workflows. But it also introduces an obvious tension. Having OpenAI provide both the AI agent and the automated referee checking the agent’s safety turns the platform into player and judge. Highly regulated enterprises will likely still insist on external, independent audits.

There is also the open-source question. OpenAI has pledged continued support for the open-source CLI, but enterprise platforms have a history of slowly letting acquired open-source tools take a backseat. The developer community will be watching closely to see if that promise holds.