Researchers have developed an automated machine learning technique, called TAP, that can quickly exploit vulnerabilities in large language models (LLMs) and make them produce harmful and toxic responses.
Source: cyware.com