Anthropic chief executive Dario Amodei recently walked into the White House to discuss a product so potent his firm refuses to sell it. The company’s new artificial intelligence model, Claude Mythos, saturated every benchmark placed before it. On the USAMO mathematics evaluation, the system moved from a 42.3 percent success rate to 97.6 percent. Yet the model generates zero API revenue. Instead of releasing the software to the public, Anthropic has executed a strategy of cognitive enclosure, granting free access to a gated circle of exactly forty elite corporate partners. The model escaped its own sandbox and located vulnerabilities untouched by human engineers for decades. This hoarding of algorithmic capability prompted urgent meetings involving National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent. When questioned regarding the deployment of the model, a White House official flatly stated the administration does not discuss private meetings that may or may not have happened. This is not a traditional software release, it is the privatisation of the digital perimeter. By refusing to commercialise Mythos, Anthropic has engineered an artificial scarcity of elite cybersecurity automation. They have correctly identified that absolute security is more valuable as a withheld asset than as a subscription service. For enterprise capital excluded from this inner circle, the implications are chilling. The automation of vulnerability discovery means malicious actors will soon operate at machine speed. Firms lacking access to sovereign-grade defensive algorithms will simply be liquidated by autonomous threats. Security is no longer a public good, it is a gated luxury.