Anthropic develops restricted Mythos successor

Anthropic's pursuit of a model more powerful than the restricted 'Mythos' reveals that frontier AI labs cannot resist the capabilities race, pushing safety-banning thresholds higher and turning yesterday's 'too dangerous' models into tomorrow's commercial baselines.

* Safety boundaries are moving targets: A model once deemed too capable to release is now being succeeded, suggesting that what was once considered unsafe is quickly normalized as frontier development continues.

* The capability race continues: Despite public emphasis on safety, Anthropic's internal roadmap remains aggressively focused on scaling raw capabilities.

* Defensive vs. offensive dual-use: Advanced models like the Mythos family raise critical concerns about whether restricted access programs (such as Project Glasswing) can successfully contain powerful capabilities without eventual leaks or commercial pressure.