Anthropic recently announced a limited release of its newest artificial intelligence model, dubbed Mythos, citing its advanced capability in identifying security exploits within software utilized globally. Rather than making Mythos publicly available, the frontier AI laboratory has opted to share it exclusively with a select group of major companies and organizations that oversee critical online infrastructure, including prominent entities such as Amazon Web Services and JPMorgan Chase. Reports suggest that OpenAI is contemplating a similar strategy for its forthcoming cybersecurity tool.

The stated objective behind this restricted dissemination is to empower these large enterprises to proactively counteract malicious actors who might leverage sophisticated large language models (LLMs) to breach secure software systems. However, industry observers hint at additional motivations beyond mere cybersecurity concerns or the conventional hyping of model capabilities. Dan Lahav, CEO of the AI cybersecurity lab Irregular, previously noted that while AI tools' ability to discover vulnerabilities is significant, the actual value of a weakness to an attacker is contingent upon various factors, particularly how vulnerabilities can be chained together. Lahav emphasized the importance of determining whether AI-discovered exploits are