OpenAI's GPT-4 can exploit real vulnerabilities by reading security advisories

While some other LLMs appear to flat-out suck

AI agents, which combine large language models with automation software, can successfully exploit real world security vulnerabilities by reading security advisories, academics have claimed....