OpenAI Rolls Out GPT-5.4-Cyber With Restricted Access; Anthropic's Mythos Escalates Security AI Race

OpenAI released its defensive security model, GPT-5.4-Cyber, on April 14 to a select group of Trusted Access for Cyber participants—a direct countermove to Anthropic's Mythos Preview announcement a week prior. The shift signals a policy pivot toward verified access over outright restrictions.