'Too Dangerous to Release' Is Becoming AI's New Normal (time.com)

OpenAI has introduced GPT-Rosalind, an advanced AI model designed for life sciences, which surpasses previous models in chemistry, biology, and experimental design. However, like Anthropic's Claude Mythos and OpenAI’s GPT-5.4-Cyber, it is not publicly accessible and is available only to “qualified customers” under a trusted access program. This decision reflects a disturbing trend where AI companies restrict access to their most powerful models due to concerns about their potential misuse, prompting discussions on whether private firms should dictate the terms of AI development and access. The restrictions stem from the dual-use implications of these technologies; the same models that aid scientific research could also be weaponized for bioterrorism or cyberattacks. While companies like OpenAI and Anthropic ensure access to organizations with robust internal controls, defining "legitimate" users poses challenges, especially beyond U.S. borders. As the race for AI capabilities accelerates, concerns grow over the potential risks associated with open-source models, which may soon equal proprietary systems in performance, raising the stakes for cybersecurity and ethical considerations surrounding AI deployment. This shift underscores the urgent need for regulatory frameworks to manage and oversee the development and use of advanced AI technologies.