Anthropic says Mythos-class model will be released “after robust safety measures” as it explains its progress

Anthropic says its Mythos-class model has found over 10,000 high-severity or critical vulnerabilities through Project Glasswing.

The company is keeping the model private for now and will share it only with select partners.

Anthropic says a public release would wait until stronger safety measures and defensive guidance are in place.

The move reflects the company’s view that powerful AI can help find bugs faster, but can also speed up exploitation.