GPT-Red: OpenAI’s AI That Attacks Its Own Models
GPT-Red, OpenAI's internal attacker AI, lands 84% of its attacks on GPT models versus 13% for humans. AI security just changed methods.
Articles tagged prompt injection
OpenAI rolls out Lockdown Mode against prompt injection
Lockdown Mode: OpenAI's new ChatGPT Business security setting disables web, images, deep research and agent mode to cut prompt injection exposure.