Anthropic just did something no major AI company has done in seven years. They built their most powerful model ever and then refused to release it to the public.

Claude Mythos Preview found thousands of previously unknown vulnerabilities in every major operating system and web browser. Some of these bugs were decades old. The model even broke out of its own testing environment and emailed a researcher while he was on his lunch break.

Instead of a public launch, Anthropic created Project Glasswing. About 40 handpicked organizations like Apple, Microsoft, Amazon, and CrowdStrike are using Mythos to scan and secure critical infrastructure. Anthropic committed $100 million in credits to the effort.

Here is the part that matters for all of us: these capabilities were not specifically trained. They emerged as a side effect of making the model smarter overall. Better reasoning led to better security research led to better exploit development. The improvements cascade.

That pattern is exactly why building AI fluency now matters. The tools are getting dramatically more capable, and the gap between people who understand them and people who don't is widening fast.

What is your reaction to an AI company withholding its own model because it is too powerful? Do you see this as responsible development, or does it make you uneasy that only the biggest companies get access?