OpenAI acquires AI security company Promptfoo: elevating security testing and red team exercises to native Frontier features

OpenAI announces acquisition of AI safety platform Promptfoo, integrating its technology into the Frontier platform to provide automated red teaming tests and vulnerability protections for enterprise AI Agent deployment.
(Background: Sam Altman’s questionable actions? Just after being blocked by the Pentagon, Anthropic shifts to support OpenAI in securing U.S. Department of Defense contracts.)
(Additional context: The Wall Street Journal reports: Trump’s targeting of Iran’s Khamenei relies on Claude AI positioning, with OpenAI taking full control of Pentagon systems.)

OpenAI announced yesterday (9th) the acquisition of AI safety platform Promptfoo, a startup founded in 2024 specializing in vulnerability testing and red team exercises for large language models (LLMs)—simulating real hacker behaviors to strengthen cybersecurity defenses.

As AI evolves from chatbots to autonomous “AI colleagues” with execution permissions, preventing these agents from jailbreaking or leaking sensitive data has become a core challenge for large-scale enterprise adoption.

Security Risks in the Era of AI Agents

According to OpenAI’s official announcement, Promptfoo’s technology will be deeply integrated into the newly launched enterprise platform OpenAI Frontier, released in February 2026, supporting companies building agents on Frontier:

• Built-in security testing: Automated security assessments and red team exercises will be native features of the Frontier platform, helping enterprises identify and fix risks such as rapid injection, jailbreaking, data leaks, tool misuse, and agent behavior beyond intended strategies.
• Embedding security and evaluation into development workflows: Frontier will be deeply integrated with workflows for identifying, investigating, and mitigating agent risks, making security a core part of AI system development and operation.
• Supervision and accountability: Reporting and traceability features will help organizations document tests, monitor changes over time, and meet the increasing governance, risk, and compliance demands of AI.

According to the announcement, over 25% of Fortune 500 companies are already using Promptfoo’s open-source tools, with 350,000 developer users. The 23-person team received $23 million in funding, and after its latest funding round in July 2025, the company was valued at $86 million.

Promptfoo founders Ian Webster and Michael D’Angelo will lead the entire team joining OpenAI.

Security and Commitment as Key Focus

In simple terms, AI Agents are gradually shifting from “information-seeking students” to “personal assistants with your stamp of approval.”

This transformation elevates risks from data leaks to loss of control over actions. When we grant AI autonomy to perform tasks, the greatest danger is no longer incorrect statements, but biases in understanding intent or being misled by hidden commands from hackers—potentially causing irreversible actions like unauthorized transfers or deleting critical files.

In environments where multiple AIs collaborate, a logical error in one agent can trigger catastrophic chain reactions.

Therefore, the core of security in the Agent era is not about blocking information but about “monitoring behavior.” We must manage AI like employees—setting clear permission boundaries and review mechanisms. Only by making AI actions transparent and permissions precisely controlled can this powerful automation become an asset rather than a backdoor that’s difficult to defend.