Anthropic decision to restrict the public release of its Mythos autonomous hacking model is being sold to the public as a triumph of corporate responsibility. In reality, it is a masterclass in enterprise rent extraction. By launching Project Glasswing, Anthropic has limited access to this powerful artificial intelligence solely to a vetted cartel of corporate partners, including Nvidia, Microsoft, and Apple. Generative AI is increasingly capable of identifying and developing exploit chains faster than human engineers can patch them. By hoarding the Mythos model, Anthropic is constructing the ultimate digital tollbooth. The biological developer class writes fundamentally flawed code; the sovereign AI detects the resulting vulnerabilities. Rather than democratising this auditing capability, tech monopolies are enclosing it behind lucrative corporate partnerships and massive usage credits. This cognitive enclosure is highly rational. It forces legacy enterprises to rely on elite, closed-source models to secure their perimeters against the inevitable cyber-threats generated by the broader market. Security is no longer a shared digital commons; it is a premium subscription service sold by the firms that engineered the threat in the first place. The transition of agentic hacking AI from the software commons to state-vetted security monopolies provides a highly predictable margin expansion for elite capital.