OpenAI is acquiring AI security startup Promptfoo to bolster security testing for enterprise AI systems and to integrate Promptfoo’s tools into OpenAI Frontier, its enterprise platform for AI coworkers.

Promptfoo has raised $23 million in funding and carried a reported July 2025 valuation of $86 million.

Promptfoo’s testing framework lets users define prompts, expected outputs, and evaluation criteria in configuration files, runs prompts across multiple models or AI apps, and scores results against predefined rules or automated grading functions.

Risk notes include Microsoft’s Piotroski F-Score of 8 and Beneish M-Score of -2.66, signaling solid financial health with low manipulation risk, while acknowledging sector-specific risks from rapid tech change and competition.

Additional risk factors cited include rapid tech changes, competitive pressures in software, a beta of 1.16 indicating moderate market volatility, and overall financial health signals.

Core capability highlighted is regression testing to compare outputs across model versions or configuration changes before releases.

Expert commentary from Dr. Elaine Chen praises proactive safety testing and notes industry-wide standards being pursued.

If testing DNA becomes default guardrails and remains interoperable, agent security could become a competitive selling point, with safety becoming a primary performance metric.

Page features sponsor content and multiple partner highlights which may influence depth and reliability of narrative.

OpenAI released an open-source evaluation framework for comparing outputs and benchmarking model performance under various conditions.

The move comes amid broader AI infrastructure competition as enterprises seek risk management, compliance, and predictable performance alongside advanced models.

Automated red-teaming resembles CI/CD for security, defining threat models and continually generating adversarial scenarios scored against policies for data boundaries and least-privilege tool use.