Prompt Injection

concepts · 2 notes linked

Related: Openai · Chatgpt · Microsoft · Adversarial Machine Learning · Large Language Models · Discord · AI Jailbreaking · AI Safety

Notes

• Can AI really be protected from text-based attacks? | TechCrunch — LLM prompt injection attacks are low-barrier and currently unpreventable
• People Are Using A 'Grandma Exploit' To Break AI - Kotaku — Roleplay persona prompts bypass AI safety guardrails