Continuously hardening ChatGPT Atlas against prompt injection attacks

3 points| forks | 1 month ago |openai.com