A new report from the California-based AI safety nonprofit FAR.AI tested the guardrails of models from four major US companies, finding that some were vulnerable to "jailbreaks" designed to trick them into generating harmful content.
register or log in to get more services such as weather report, extended news articles and web search