Anthropic has restricted the public release of its latest advanced model, Mythos, citing concerns over its exceptional ability to discover software security vulnerabilities. Rather than making the model widely available, the company plans to grant access only to a select group of major enterprises and organizations that manage critical digital infrastructure, including Amazon Web Services and JPMorgan Chase. This strategic move aims to help these entities proactively defend against cyber threats posed by advanced large language models. Reports indicate that OpenAI is considering a similar approach for its upcoming cybersecurity tools. The stated justification focuses on cybersecurity safety. Anthropic claims Mythos can identify and exploit software flaws far more effectively than its previous Opus model. However, some industry experts argue that the necessity of such strict limitations may extend beyond genuine security risks. Dan Lahav, CEO of the AI cybersecurity lab Irregular, noted that the true danger of an AI-discovered vulnerability depends on its specific context and potential for exploitation in combination with other factors. Furthermore, independent researchers suggest Mythos may not be as uniquely powerful as claimed. The AI cybersecurity startup Aisle reported replicating similar results using smaller, open-weight models, implying that model capability for this task is not solely dependent on massive parameter counts. A significant alternative theory suggests the restriction is driven by commercial interests rather than pure safety. By limiting access to large enterprise contracts, Anthropic may be protecting a lucrative revenue stream while preventing competitors from using a technique called distillation. Distillation involves using a powerful frontier model to train smaller, cheaper models, a strategy that could allow smaller labs to challenge the market dominance of major players. David Crawshaw, CEO of exe.dev, argued that gating top-tier models behind enterprise agreements serves as a marketing shield to keep the market focused on high-value contracts. He warned that once such restrictions are in place, public access is delayed indefinitely as companies push for the next iteration, maintaining the flow of enterprise funding. The AI ecosystem is increasingly defined by this divide. Frontier labs race to develop the most capable models, while companies like Aisle seek economic advantage by utilizing multiple models and open-source alternatives, some of which are allegedly developed through distillation. While Anthropic has not explicitly confirmed whether distillation concerns influenced their decision, the timing and scope of the Mythos release suggest a dual strategy. The company may be simultaneously attempting to protect the internet's security infrastructure and safeguarding its own commercial dominance by keeping its most advanced technology exclusive to large-scale partners. Whether Mythos poses an unprecedented threat to global software security or serves primarily as a competitive moat remains to be fully determined, but the decision reflects a cautious, yet potentially profit-driven, approach to deploying frontier AI capabilities.