Promptfoo raises $18.4M to build AI security and evaluation platform
Promptfoo Inc., a broadly adopted open-source framework for artificial intelligence evaluation and testing, today announced it has raised $18.4 million in an early-stage funding led by global software investor Insight Partners.
The company’s existing investor Andreessen Horowitz also participated in the Series A round.
Founded in 2024, Promptfoo provides robust AI security testing and risk management services using standards that include red teaming, guardrails and continuous monitoring.
Red teaming is a security practice that involves simulating adversarial attacks against an AI application or model to identify vulnerabilities and weaknesses using prompts that could be exploited by attackers. It is a proactive approach that goes beyond traditional testing methods by evaluating numerous portions of the AI stack, including model behavior, data pipelines and user interactions.
The term itself originally comes from cybersecurity, where it is the job of the “red team” to behave like attackers so that defenders can learn to protect software and computer networks better. This practice enables the development of more effective guardrails for AI, allowing companies like Promptfoo to generate smarter defenses that are better equipped to handle more sophisticated attacks.
Recent headlines have outlined the need for increased AI model security, such as when Elon Musk’s X Corp.’s Grok chatbot went rogue earlier this month, making antisemitic remarks and naming itself “MechaHitler.” Public AI meltdowns are only part of enterprise concerns, as attackers can use AI models to leak sensitive information, run restricted code or allow malware to infiltrate networks with carefully crafted language.
“The internet is being rebuilt for agents and the APIs they call are increasingly being exposed as Model Context Protocol servers,” Ian Webster, chief executive and co-founder of Promptfoo told SiliconANGLE. “This coincides with a shift from single-turn text to long-running, tool-using agents that can collaborate, hand off tasks, and critique each other. These agents and tools may live in untrusted environments or may execute instructions from untrusted tools, may write and execute code, and can have access to sensitive data.”
Model Context Protocol, or MCP, is an open protocol recently released by Anthropic PBC that standardizes how AI applications talk to large language models. It serves as a universal means for AI agents to gather information from various sources and interact with tools across the internet. Due to its popularity, Promptfoo has added a tool that proxies and scans data passed between AI and external sources, checking it for potential compliance and security issues.
Webster added that enterprises are beginning to adopt new technologies such as retrieval-augmented generation, placing sensitive documents into the system, AI agents and the above-mentioned MCP.
“These new architectures introduce unique risk,” Webster said.
The company embeds its automated red team tests directly into the build and release process, pairing findings with clear, actionable remediation guidance for AI security teams to manage vulnerabilities and risks.
This security spans attack objective generation, exploit search and regression testing. Security teams have a single, continuous workflow for hardening their applications from simple chatbots to complex agentic systems. Adaptive guardrails and monitoring can prevent risks such as prompt injection, data leakage and insecure tool use.
“Our long‑term vision is for Promptfoo to play the same role in AI infrastructure that continuous integration pipelines play in DevOps: every new model or prompt change, or tool integration is automatically evaluated, red‑teamed and either blocked or promoted with a signed safety report,” Webster said.
The company said its open-source tools have been used by over 100,000 developers and more than 30 Fortune 500 companies.
