# OpenAI Reveals How ChatGPT Defends Against Prompt Injection Attacks

OpenAI announced new details about how it designs AI agents to resist prompt injection and social engineering attacks, focusing on protecting ChatGPT users from malicious manipulation.

The company explained that its defense strategy centers on two key approaches: constraining risky actions that AI agents can take and protecting sensitive data within agent workflows. Prompt injection attacks occur when malicious actors try to trick AI systems into ignoring their original instructions and following harmful commands instead.

This matters because AI agents are increasingly being given access to sensitive information and the ability to perform actions on behalf of users—from accessing emails to making purchases. Without proper safeguards, a cleverly worded prompt could potentially manipulate an AI into leaking private data or performing unauthorized actions.

OpenAI's announcement signals growing industry awareness of AI security vulnerabilities as these systems become more powerful and autonomous. The company