Overview
Severity: HIGH | Affected: Multiple LLM Providers | Category: research
A new paper published by researchers at the Stanford AI Lab details a novel jailbreak technique named 'Chrono-Prompting'. The method bypasses the safety filters of leading large language models by framing harmful requests within complex temporal contexts. By asking the model to generate content as if it were a historical document from a different era or a speculative script from a future scenario, the researchers were able to circumvent ethical guards with over an 85% success rate across multiple tested models. The technique exploits inconsistencies in how models apply safety rules across different perceived timelines. The findings demonstrate a sophisticated new vector for adversarial attacks and prompted calls for AI developers to strengthen models' temporal reasoning and contextual safety alignment.