CHATGPT JAILBREAK: RESEARCHERS BYPASS AI SAFEGUARDS USING HEXADECIMAL ENCODING AND EMOJIS

Summary from the TLDR InfoSec newsletter

ChatGPT's safety controls can be bypassed using malicious instructions encoded in hexadecimal format. The vulnerability was disclosed through Mozilla's 0Din bug bounty program. It represents another example of prompt injection bypassing AI safeguards.