Apr 09, Kathmandu - OpenAI, the leading company behind ChatGPT, is finalizing the development of a new, highly advanced AI model with robust cybersecurity capabilities. The company plans to limit access to this model, initially restricting it to select organizations.

According to Axios, this move mirrors Anthropic’s recent limited public deployment of its Mythos model, emphasizing a cautious approach to releasing powerful AI technologies.

Experts warn that AI systems with autonomous decision-making and hacking abilities pose significant security risks. Manufacturers are increasingly hesitant to release such potent models openly, fearing potential misuse or destructive applications.

Anthropic announced on Tuesday that it would restrict access to its new Mythos Preview model to a few selected cybersecurity firms, marking the first time an AI company has adopted such an approach. OpenAI is reportedly considering a similar strategy.

Following the release of its most powerful cybersecurity-capable model, GPT-5.3-Codex, in February, OpenAI launched a pilot program called Trusted Access for Cyber, granting authorized institutions access to more powerful models for security purposes.  

Security experts have been warning for the past year that these AI models, if misused, could autonomously disrupt essential services such as water supply, electricity, or financial systems. It appears these capabilities are now becoming a reality.

Despite efforts by AI firms to restrict access, security analysts emphasize that there is no turning back. Rob T. Lee, Chief AI Officer at SANS Institute, stated, “You cannot prevent AI models from discovering code vulnerabilities. This capability already exists.”

Wendy Whitmore of Palo Alto Networks predicts that similar open models with such capabilities will appear in the market within weeks or months. Adam Myers of CrowdStrike calls Mythos’s potential a “warning bell for the entire industry.”

Experts agree that a phased approach to releasing these advanced AI models is essential. While Anthropic has stated it will never make Mythos publicly available, OpenAI has not yet clarified whether its upcoming models will be widely released or kept restricted.

Meanwhile, researchers warn that many publicly available AI models today already possess vulnerabilities that could be exploited.