OpenAI is finalizing a new model with advanced cybersecurity capabilities that it plans to release only to a small group of companies due to concerns about its advanced hacking capabilities. This move comes after Anthropic announced plans to limit access of its new Mythos Preview model to a select group of tech and cybersecurity companies due due to fears of its advanced capabilities. The company, which introduced its "Trusted Access for Cyber" pilot program in February, provided participants with access to more cyber-capable models to accelerate defensive work. However, it is unclear if OpenAI will decide to release its new model more broadly. Top security experts warn that there is no going back, as AI models are already capable of finding vulnerabilities and exploits that could disrupt water utilities, the electric grid, or financial systems.