“I Had a Dream” and Generative AI Jailbreaks
Chatbots can be manipulated by attackers to yield malicious outputs, such as code for malware or illegal activities, through “prompt engineering” and “prompt injections”. Cybersecurity researchers have created a ‘Universal LLM Jailbreak’ that can thwart restrictions of major AI systems like ChatGPT, Google Bard, Microsoft Bing, and Anthropic Claude. As more organizations integrate these systems, susceptibility to these attacks rises. Although developers frequently update models, identifying malicious prompts remains challenging. Establishing trust boundaries and implementing security guardrails are necessary preventive steps.
Source: thehackernews.com
- Read more