Researchers at Hacktron used Anthropic’s latest Claude model to exploit a vulnerability in OpenAI’s systems, highlighting the growing cyber capabilities of AI tools.