# OpenAI Reveals How It Monitors AI Coding Agents for Safety Risks

OpenAI announced a new approach to detecting potential safety issues in its internal AI coding agents, using a technique called "chain-of-thought monitoring."

The AI research company shared that it's actively analyzing how its coding agents think and operate in real-world deployments. The goal is to identify any signs of misalignment—when an AI system's behavior doesn't match its intended purpose or could pose risks.

Chain-of-thought monitoring works by examining the reasoning process AI agents use when writing code. This allows OpenAI's safety teams to spot concerning patterns before they become problems, rather than only evaluating final outputs.

**Why it matters:** As AI coding assistants become more powerful and autonomous, ensuring they remain aligned with human intentions is critical. Misaligned AI agents could potentially write malicious code, bypass security measures, or pursue unintended goals