OpenAI Makes Two Moves at Once: Security + Capability
This week, OpenAI announced two major developments that clearly signal the company is seriously preparing for the era of enterprise-grade Agentic AI.
1. Acquiring Promptfoo — AI Red-Teaming Becomes Built-In
Promptfoo is an open-source platform for testing AI security, already used by more than 25% of Fortune 500 companies. It covers everything from prompt injection testing and jailbreak detection to data leak prevention.
OpenAI announced the acquisition of Promptfoo, with Promptfoo remaining open source under the Apache 2.0 license, while also being integrated into the OpenAI platform as a built-in security layer.
What enterprises will gain:
- Red-teaming as a service: Test AI systems before deployment to identify vulnerabilities
- Automated vulnerability scanning: Automatically scan for prompt injection, data exfiltration, and harmful output
- Compliance reporting: Generate security reports for auditors and regulators
2. GPT-5.4 — AI That Can Use a Computer on Its Own
GPT-5.4 launches with a game-changing feature: native computer-use capabilities. This enables AI agents to:
- See the screen through screenshots
- Click the mouse at the required position
- Type text into applications
- Open and close apps and navigate between windows
Imagine this: instead of building API integrations for every system, an AI agent can simply “sit in front of the screen” and work like an employee using a computer—opening the ERP, entering data, exporting reports, and sending emails, all without writing a single connector.
