A recent investigation by the Financial Times highlights the alarming emergence of software tools that can quickly bypass safety measures in open-source AI models. One such tool, Heretic, can remove guardrails from models like Google’s Gemma 3 and Meta’s Llama 3.3 in under ten minutes, allowing them to generate dangerous content, including instructions for conducting chlorine gas attacks and creating viruses for credit card theft. Heretic, which is freely available on GitHub, has been used to create over 3,500 modified models, downloaded 13 million times. Experts stress that this ease of access to dangerous capabilities poses significant risks, as average users can now exploit these technologies without extensive technical knowledge. While proprietary models like OpenAI's ChatGPT are currently safe from such tools, the potential for misuse in open-source models is a growing concern. Google has acknowledged the risks associated with such tools and claims to implement rigorous safety evaluations for its models.