OpenAI moves to acquire Promptfoo to deepen agentic security testing

What OpenAI announced

On March 9, 2026, OpenAI said on X that it is acquiring Promptfoo. The company framed the deal as a direct investment in the security and evaluation layer around agentic systems, saying Promptfoo's technology will strengthen agentic security testing and evaluation inside OpenAI Frontier.

That wording matters because a growing share of AI risk no longer sits only in raw model output. It sits in tool use, permission boundaries, indirect prompt injection, system prompts, and whether a team can repeatedly test how an agent behaves before it reaches production. OpenAI is signaling that this testing stack is becoming core infrastructure rather than a side utility.

What stays the same for Promptfoo users

OpenAI's public note also included two concrete commitments. First, it said Promptfoo will remain open source under the current license. Second, it said OpenAI will continue to service and support current customers. Those are important details because acquisitions in developer tooling often create uncertainty around licensing, product continuity, and enterprise support.

The public note from Promptfoo's team points in the same direction. It describes Promptfoo as software developers already use to evaluate, red-team, and secure LLM and agentic systems, and says the team is joining OpenAI rather than winding the project down. In practical terms, OpenAI is not only buying a codebase. It is bringing in a workflow that many AI teams already use to measure and harden real deployments.

Why this matters

This move fits a broader shift in the AI stack. As access to strong models becomes more common, the harder product problem becomes reliability under real workflows. Enterprises do not just need better models; they need repeatable ways to catch jailbreaks, unsafe tool invocation, evaluation drift, and regressions across fast-moving agent deployments.

By bringing Promptfoo into OpenAI Frontier, OpenAI is effectively treating evals and security testing as strategic product surfaces. The practical open question is how much of Promptfoo's workflow remains broadly usable outside OpenAI's own platform over time. But the direction is already clear: security evaluation is moving closer to the center of agent platform development.

Sources: OpenAI X post · OpenAI: OpenAI to acquire Promptfoo · Promptfoo blog