OpenAI Acquires Promptfoo: AI Agent Security Takes Center Stage
OpenAI has acquired Promptfoo. The tool security researchers prefer for testing language model vulnerabilities now enters the ecosystem of ChatGPT's creator.
The acquisition took place in April 2026. Financial terms haven't been disclosed.
What Promptfoo Is
Promptfoo is an open-source tool for LLM red teaming. It simulates attacks — prompt injection, jailbreaking, data leakage — and measures a model's resistance before a malicious actor does.
Before the acquisition: over 5,000 stars on GitHub. Used by security teams at hundreds of companies. A reference point in the field of systematic AI risk assessment.
The main features:
- Prompt injection testing — checks whether external input can hijack the model's instructions
- Jailbreak scanning — tests hundreds of guardrail circumvention techniques
- Data leakage detection — identifies when the model exposes data it shouldn't
- Automated red teaming — automatically generates attack scenarios based on the application context
Why OpenAI Bought Promptfoo
OpenAI is expanding its lineup of autonomous agents. Operator, Canvas, Deep Research: systems that act in the real world, access sensitive data, and carry out irreversible operations.
An agent vulnerable to prompt injection isn't a theoretical risk. It's an operational one.
The Promptfoo acquisition brings three things in-house for OpenAI:
- Testing technology — methodologies already validated across thousands of real use cases
- A specialized team — security researchers focused exclusively on LLMs
- Positioning — a signal to the market: security as a priority, not an afterthought
The Context: AI Agent Security Is 2026's Big Problem
The problem isn't new. It's urgent.
With agents booking flights, sending emails, managing code, and accessing corporate CRMs, the attack surface has multiplied. A single malicious prompt in a document processed by an agent can compromise an entire session.
Anthropic, Google, and Microsoft are investing heavily in the same space. OpenAI, with Promptfoo, is accelerating on a front where competitive advantage is measured in enterprise trust.
Companies choose their AI provider partly based on who offers the strongest security guarantees. Not just who offers the most capable model.
What Changes for Developers
Promptfoo will remain open source in the short term — it's part of the value of the acquisition. But integration into the OpenAI ecosystem will likely bring:
- Native API testing — security evaluations built directly into development tools
- Enterprise certifications — automatic audits for use in regulated sectors (finance, healthcare, legal)
- Continuous red teaming — monitoring in production, not just during development
FAQ
What is red teaming in AI systems? AI red teaming is the practice of simulating systematic attacks on a language model to identify its vulnerabilities before deployment. It includes prompt injection, jailbreaking, sensitive data extraction, and out-of-spec behavior.
What exactly does Promptfoo do? Promptfoo automates LLM security evaluation. It generates attack scenarios, runs them against the target model, classifies the vulnerabilities found, and produces structured reports for security teams.
Will OpenAI keep Promptfoo open source? In the short term, yes. Open source is part of the product's value. Enterprise features and integration with the OpenAI API will likely be monetized separately.
Which companies used Promptfoo before the acquisition? Promptfoo was used by security teams at AI startups, banks, healthcare companies, and big tech. Any organization deploying LLMs in production considered it part of its security toolchain.