Frontier AI labs' voluntary delays are rewriting the regulatory playbook

Anthropic's deferral of Claude Mythos—its cybersecurity-focused model—is now part of active European Commission review, illustrating a tightening feedback loop: voluntary company decisions on capability deployment are folding directly into regulatory conversations. OpenAI, Google DeepMind, and Anthropic have all adjusted release timelines in response to government engagement. The structural shift: labs are no longer simply releasing and then facing rules, but negotiating capability access upstream. Whether this constitutes genuine safety governance or regulatory capture remains contested.